Basic Algebraic Geometry

Grundlehren der
mathematischen Wissenschaften 213

A Series of Comprehensive Studies in Mathematics
Editors
S. S. Chern J. L. Doob J. Douglas,jr.
A. Grothendieck E. Heinz F. Hirzebruch E. Hopf
S. Mac Lane W. Magnus M. M. Postnikov
W Schmidt D. S. Scott K. Stein J. Tits
B. L. van der Waerden
Managing Editors
B.Eckmann J.K.Moser
I. R. Shafarevich
Basic
Algebraic Geometry
Translated from the Russian by K. A. Hirsch
With 19 Figures
Revised Printing
Springer-Verlag
Berlin Heidelberg New York 1977
Igor R. Shafarevich
Steklov Mathematical Institute of the Academy of Sciences of the USSR
Title of the Russian Original Edition: Osnovy algebraicheskoi geometrii,

Publisher: Nauka, Moscow, 1972
AMS Subject Classifications (1970): Primary 14-XX, Secondary 20 GXX
ISBN-13: 978-3-540-08264-4 e-ISBN-13: 978-3-642-96200-4

DOl: 10.1007/978-3-642-96200-4
This work is subject to copyright. All rights are reserved, whether the whole or part
of the material is concerned, specifically those of translation, reprinting, re-use of
illustration, broadcasting, reproduction by photocopying machine or similar means,
and storage in data banks. Under § 54 of the German Copyright Law where copies are
made for other than private use, a fee is payable to the publisher, the amount of the
fee to be determined by agreement with the publisher.
© by Springer-Verlag Berlin Heidelberg 1974, 1977.
Solkover reprint of the hardcover I st edition 1977
Library of Congress Catalog Card Number 74-1653.
2141/3140-543210.
Preface
Algebraic geometry occupied a central place in the mathematics of the

last century. The deepest results of Abel, Riemann, Weierstrass, many of
the most important papers of Klein and Poincare belong to this do-
mam.
At the end of the last and the beginning of the present century the
attitude towards algebraic geometry changed abruptly. Around 1910
Klein wrote:
"When I was a student, Abelian functions*-as an after-effect of
Jacobi's tradition-were regarded as the undIsputed summit of mathe-
matics, and each of us, as a matter of course, had the ambition to forge
ahead in this field. And now? The young generation hardly know what
Abelian functions are." (Vorlesungen tiber die Entwicklung der Mathe-
matik im XIX. Jahrhundert, Springer-Verlag, Berlin 1926, Seite 312).
The style of thinking that was fully developed in algebraic geometry
at that time was too far removed from the set-theoretical and axio-
matic spirit, which then determined the development of mathematics.
Several decades had to lapse before the rise of the theory of topolo-
gical, differentiable and complex manifolds, the general theory of fields,
the theory of ideals in sufficiently general rings, and only then it
became possible to construct algebraic geometry on the basis of the
principles of set-theoretical mathematics.
Around the middle of the present century algebraic geometry had
undergone to a large extent such a reshaping process. As a result, it
can again lay claim to the position it once occupied in mathematics.
The range of applicability of its ideas enlarged extraordinarily towards
algebraic varieties over arbitrary fields and complex manifolds of the
most general kind. Algebraic geometry, quite apart from many better
achievements, succeeded in removing the charge of being "incompre-
hensible" and "unconvincing".
The basis for this rebuilding of algebraic geometry was algebra.
In its first versions the use of a delicate algebraic apparatus often led
* From the present-day point of view. the theory of Abelian functions is the analytical
aspect of the theory of projective algebraic group varieties.
VI Preface
to the disappearance of that vivid geometric style that was charac-

teristic for the preceding period. However, the last two decades have
brought many simplifications in the foundations of algebraic geometry,
which have made it possible to come remarkably close to the ideal
combination of logical transparency and geometrical intuitiveness.
The aim of the book is to set forth the elements of algebraic

geometry to a fairly wide extent, so as to give a general idea of this
branch of mathematics and to provide a basis for the study of the
more specialist literature. The reader is not assumed to have any
prior knowledge of algebraic geometry, neither of its general theorems
nor of concrete examples. For this reason, side by side with the
development of the general theory, applications and special cases take
a prominent place, because they motivate new concepts and problems
to be raised.
It seems to me that the logic of the subject will be clearer to the
reader if in the spirit of the "biogenetic law" he repeats, in a very
condensed way, the evolution of algebraic geometry. Therefore the very
first section, for example, is devoted to the simplest properties of
plane algebraic curves. Similarly, Part One of the book discusses only
algebraic varieties situated in a projective space, and it is only in
Part Two that the reader comes across schemes and the general con-
cept of a variety.
Part Three is concerned with algebraic varieties over the complex
field and their connections with complex analytic manifolds. In this part
the reader needs some acquaintance with the elements of topology
and the theory of analytic functions.
My sincere thanks are due to all who have helped me with their
advice during the work on this book. It is based on notes of some
courses I have given at the University of Moscow. Many members of
the audience and many readers of these notes have made very useful
comments to me. I am particularly indebted to the editor, B. G.
Moishezon. Numerous conversations with him were very useful for me.
A number of proofs spread over the book are based on his advice.
Table of Contents
Part I. Algebraic Varieties in a Projective Space

Chapter I. Fundamental Concepts
§1. Plane Algebraic Curves . 3
1. Rational Curves 3
2. Connections with the Theory of Fields. 8
3. Birational Isomorphism of Curves 11
Exercises . . . . . . . . . . 13
§2. Closed Subsets of Affine Spaces 14
1. Definition of Closed Subset . 14
2. Regular Functions on a Closed Set 16
3. Regular Mappings 18
Exercises . . . . 21
§ 3. Rational Functions 22
1. Irreducible Sets 22
2. Rational Functions 24
3. Rational Mappings 25
Exercises . 30
§ 4. Quasiprojective Varieties 30
1. Closed Subsets of a Projective Space 30
2. Regular Functions 33
3. Rational Functions 38
4. Examples of Regular Mappings 40
Exercises . . . . . . . . . . . 41
§ 5. Products and Mappings of Quasiprojective Varieties 41
1. Products . . . . . . . . . . . . . . . 41
2. Closure of the Image of a Projective Variety 44
3. Finite Mappings 47
4. Normalization Theorem 52
Exercises . 52
§ 6. Dimension 53
1. Definition of Dimension 53
2. Dimension of an Intersection with a Hypersurface 56
3. A Theorem on the Dimension of Fibres 60
4. Lines on Surfaces . . . . . 62
5. The Chow Coordinates of a Projective Variety 65
Exercises . . . . . . . . . . . . . . . . . 69
VIn Table of Contents
Chapter II. Local Properties

§ 1. Simple and Singular Points 71
1. The Local Ring of a Point 71
2. The Tangent Space 72
3. Invariance of the Tangent Space 74
4. Singular Points 77
5. The Tangent Cone 79
Exercises . . . . . 80
§ 2. Expansion in Power Series 81
1. Local Parameters at a Point 81
2. Expansion in Power Series 84
3. Varieties over the Field of Real and the Field of Complex Numbers 88
Exercises . . . . . . . . 89
§ 3. Properties of Simple Points 90

I. Subvarieties of Codimension 1 90
2. Smooth Subvarieties 93
3. Factorization in the Local Ring of a Simple Point 94
Exercises . . . . . . . . . . . . . . 97
§ 4. The Structure of Birational Isomorphisms 98
1. The (J"-Process in a Projective Space 98
2. The Local (J"-Process . . . . . . . . 100
3. Behaviour of Subvarieties under a (J"-Process 103
4. Exceptional Subvarieties . . . . . . . . 104
5. Isomorphism and Birational Isomorphism 105
§ 5. Normal Varieties 109
1. Normality 109
2. Normalization of Affine Varieties 113
3. Ramification 115
4. Normalization of Curves . . . . 120
5. Projective Embeddings of Smooth Varieties 123
Exercises . . . . . . . . . . . . . . . . 126
Chapter III. Divisors and Differential Forms

§ 1. Divisors . . . . . . . . . 127
1. Divisor of a Function 127
2. Locally Principal Divisors 131
3. How to Shift the Support of a Divisor Away from Points 134
4. Divisors and Rational Mappings 135
5. The Space Associated with a Divisor 137
§ 2. Divisors on Curves 140

1. The Degree of a Divisor on a Curve 140
2. Bezout's Theorem on Curves 144
3. Cubic Curves . . . . . . 145
4. The Dimension of a Divisor 146
Exercises . . . . . . . . . 147
Table of Contents IX
§ 3. Algebraic Groups . . . . . . . . . . . . 148

1. Addition of Points on a Plane Cubic Curve 148
2. Algebraic Groups . . . . . . . . 150
3. Factor Groups. Chevalley's Theorem 151
4. Abelian Varieties 152
5. Picard Varieties 153
§ 4. Differential Forms . . . . . . . . . . . . . . . . 156

1. One-Dimensional Regular Differential Forms 156
2. Algebraic Description of the Module of Differentials 159
3. Differential Forms of Higher Degrees 161
4. Rational Differential Forms 163
Exercises . . . . . . . . . . . . . . 165
§ 5. Examples and Applications of Differential Forms 166

1. Behaviour under Mappings . . . . . . 166
2. Invariant Differential Forms on a Group 168
3. The Canonical Class 170
4. Hypersurfaces 171
5. Hyperelliptic Curves 175
6. The Riemann-Roch Theorem for Curves 176
7. Projective Immersions of Surfaces 178
Exercises . . . . . . . . . . . . . . . 180
Chapter IV. Intersection Indices

§ 1. Definition and Basic Properties 182
1. Definition of an Intersection Index 182
2. Additivity of the Intersection Index 185
3. Invariance under Equivalence . . 187
4. End of the Proof of Invariance 191
5. General Definition of the Intersection Index 194
Exercises . . . . . . . . . . . . . . . . 197
§ 2. Applications and Generalizations of Intersection Indices . . . . . 198

1. Bezout's Theorem in a Projective Space and Products of Projective
Spaces . . . . . . . . . . . . . . . . 198
2. Varieties over the Field of Real Numbers . 199
3. The Genus of a Smooth Curve on a Surface 202
4. The Ring of Classes of Cycles 206
Exercises . . . . . . . . . . . 207
§ 3. Birational Isomorphisms of Surfaces 208

1. ()'-Processes of Surfaces 208
2. Some Intersection Indices 209
3. Elimination of Points of Indeterminacy 210
4. Decomposition into (),·Processes 212
5. Notes and Examples 214
Exercises . . . . . . . . . . . 216
x Table of Contents
Part II. Schemes and Varieties

Chapter V. Schemes
§1. Spectra of Rings . . . . . . . . . . . 223
1. Definition of a Spectrum . . . . . . 223
2. Properties of the Points of a Spectrum 226
3. The Spectral Topology 228
4. Irreducibility, Dimension 230
Exercises 233
§2. Sheaves 234
1. Preshea ves 234
2. The Structure Presheaf 235
3. Sheaves . . . . . 238
4. The Stalks of a Sheaf 241
Exercises 242
§ 3. Schemes 242
1. Definition of a Scheme 242
2. Pasting of Schemes 246
3. Closed Subschemes 248
4. Reducibility and Nilpotents 250
5. Finiteness Conditions 252
Exercises . . . . . 253
§ 4. Products of Schemes 254
1. Definition of a Product 254
2. Group Schemes 257
3. Separation 258
Chapter VI. Varieties

§ 1. Definition and Examples 263
1. Definitions . . . . 263
2. Vector Bundles 268
3. Bundles and Sheaves 270
4. Divisors and Line Bundles 277
Exercises . . . . . . . . . 281
§ 2. Abstract and Quasiprojective Varieties 282
I. Chow's Lemma . . . ..... . 282
2. The <1-Process along a Subvariety 283
3. Example ofa Non-Quasiprojective Variety 287
4. Criteria for Projectiveness 292
Exercises . . . . . 293
§ 3. Coherent Sheaves 294

1. Sheaves of Modules 294
2. Coherent Sheaves 298
3. Devissage of Coherent Sheaves 300
4. The Finiteness Theorem 304
Exercises . . . . . . . . . . . 305
Table of Contents XI
Part III. Algebraic Varieties over the Field of Complex

Numbers and Complex Analytic Manifolds
Chapter VII. Topology of Algebraic Varieties
§ 1. The Complex Topology 309
1. Definitions . . . . . . . . . 309
2. Algebraic Varieties as Differentiable Manifolds. Orientation 311
3. The Homology of Smooth Projective Varieties 313
Exercises . . . 318
§ 2. Connectedness 318
1. Auxiliary Lemmas 319
2. The Main Theorem 320
3. Analytic Lemmas 322
Exercises . . . . . . 324
§ 3. The Topology of Algebraic Curves 325

1. The Local Structure of Morphisms 325
2. Triangulation of Curves. . . . . 327
3. Topological Classification of Curves 329
4. Combinatorial Classification of Surfaces 333
§ 4. Real Algebraic Curves 336

1. Involutions 337
2. Proof of Harnack's Theorem 338
3. Ovals of Real Curves 340
Exercises . . . . . . . . . 341
Chapter VIII. Complex Analytic Manifolds

§ 1. Definitions and Examples 343
1. Definition . . . . . 343
2. Factor Spaces 345
3. Commutative Algebraic Groups as Factor Spaces 348
4. Examples of Compact Analytic Manifolds that are not Isomorphic to
Algebraic Varieties 351
5. Complex Spaces . . . . . . . . 357
Exercises . . . . . . . . . . . . 359
§ 2. Divisors and Meromorphic Functions 360
1. Divisors . . . 360
2. Meromorphic Functions 362
3. Siegel's Theorem 365
Exercises . . . . . . . 368
§ 3. Algebraic Varieties and Analytic Manifolds 369
1. Comparison Theorem 369
2. An Example of Non-Isomorphic Algebraic Varieties that are
Isomorphic as Analytic Manifolds . . . . . . . . . . . . . . 372
3. Example of a Non-Algebraic Compact Manifold with the Maximal
Number of Independent Meromorphic Functions . 375
4. Classification of Compact Analytic Surfaces . 377
Exercises . . . . . . . . . . . . . . . . . . . . 378
XII Table of Contents
Chapter IX. Uniformization

§ 1. The Universal Covering .......... . 380
1. The Universal Covering ofa Complex Manifold 380
2. Universal Coverings of Algebraic Curves 382
3. Projective Embeddings of Factor Spaces 384
Exercises . . . . . . . 386
§ 2. Curves of Parabolic Type 386

1. {I-Functions . . . . . 386
2. Projective Embedding 388
3. Elliptic Functions, Elliptic Curves, and Elliptic Integrals 389
Exercises . . . . . . . . 392
§ 3. Curves of Hyperbolic Type . 393

1. Poincare Series . 393
2. Projective Embedding . 395
3. Algebraic Curves and Automorphic Functions . 398
Exercises . . . . . . . . . . . . . . . . . . 400
§ 4. On the Uniformization of Manifolds of Large Dimension . 401

1. Simple Connectivity of Complete Intersections . 401
2. Example of a Variety with a Preassigned Finite Fundamental
Group .402
3. Notes 406
Exercises 408
Bibliography 409
Historical Sketch . . . . . . 411
1. Elliptic Integrals 412
2. Elliptic Functions .413
3. Abelian Integrals 415
4. Riemann Surfaces 417
5. The Inversion Problem 419
6. Geometry of Algebraic Curves 421
7. Many-Dimensional Geometry 423
8. The Analytic Theory of Manifolds 426
9. Algebraic Varieties over an Arbitrary Field. Schemes. 428
Bibliography for the Historical Sketch 431
Subject Index 433
List of Notation . 438
Advice to the Reader
The first two parts of the book assume very little knowledge on the
part of the reader. This amounts to the contents of a university course
in algebra and analytic geometry and the rudiments of the theory of
fields, which the reader could find, for example, in any of the following
books: van der Waerden, Algebra, Vol. I, Ch. V and VIII, Zariski
and Samuel, Commutative algebra, Vol. I, Ch. II, or Lang, Algebra,
Ch. VII and X. Apart· from this, frequent use is made of Hilbert's
Basis Theorem and Hilbert's Nullstellensatz. Proofs are contained in
the book by Zariski and Samuel, Vol. I, Ch. IV, § 1 and Vol. II,
Ch. II, §3.
In addition, in a few places we use certain isolated results of
commutative algebra, and for their proofs the reader is referred to the
book by Zariski and Samuel. In all cases the matter concerns only a
few pages, which can be understood independently of the remaining
parts of the book.
The third part ·assumes more knowledge. Essentially this concerns
topology. Singular homology theory is taken as known, as are properties
of differential forms, and Stokes' theorem. In Ch. VII, § 1 the concept
of a differentiable manifold is applied, also Poincare's duality law,
and some properties of intersections of cycles; in §§ 3 and 4 of the
same chapter we use the combinatorial classification of surfaces, but
these three sections are not necessary for an understanding of the rest
of the book. In the last section of the book we use one result from
Morse theory, which can, however, simply be taken on trust. Finally,
in the third part the reader is assumed to be familiar with the elements
of the theory of analytic functions-a standard university course is
amply sufficient.
The second and third parts of the book are based on the first.
However, there are passages in it that are not needed for an understanding
of what follows. They are Ch. IV, §§ 2 and 3, Ch. I, §§ 6.4 and 6.5,
Ch. II, § 1.5, Ch. II, § 4.5, Ch. II, § 5.5, Ch. III, §§ 5.6 and 5.7; Ch. III, § 3
is fairly isolated: it is connected only with Ch. VIII, § 1.3.
XIV Advice to the Reader
The reader who is interested only in vanetles over the field of

complex numbers, and altogether in the more "classical" aspects of
algebraic geometry, might study Ch. V only superficially. Finally,
there are a number of places where we report without proofs on
further developments of the questions considered in the book. Of
course, these passages are not essential for an understanding of what
follows.
In conclusion I wish to indicate the literature that has a bearing
on the problems treated in the book and can form the nucleus of a
further intensive study of algebraic geometry.
Every reader who is interested in algebraic geometry simply has to
study the cohomology theory of algebraic coherent sheaves. Within
the framework of the theory of varieties one can become acquainted
with it through Serre's paper "Faisceaux algebriques coherents", Ann.
of Math. (2) 61, 197-278 (1955), or Zariski's "Algebraic sheaf theory",
Bull. Amer. Math. Soc. 62, 117-141 (1956). Within the framework of the
theory of schemes there is an account of the theory in the notes of Manin's
"Lectures on algebraic geometry", Moscow State University 1968.
A natural continuation of this theory is the general Riemann-Roch
theorem, which can be read up in the paper by Borel and Serre, "Le
theoreme de Riemann-Roch", Bull. Soc. Math. France 86, 97-136
(1958), or Manin's "Lectures on the K-functor in algebraic geometry"
[Uspekhi Mat. Nauk 24:5, 3-86 (1969) = Russian Math. Surveys 24:5,
1-89 (1969)].
In this book there are frequent references to the Riemann-Roch
theorem for curves, but it is never proved. Of course, it follows from
the general Riemann-Roch theorem, but it can also be easily derived
directly from general properties of the cohomology of algebraic
coherent sheaves. Such a proof can be found in the book by Serre
"Groupes algebriques et corps de classes", Hermann et Cie., Paris 1959,
Ch. II. One can become acquainted with the theory of algebraic surfaces
in the book "Algebraic surfaces", Trudy Mat. Inst. Steklov 75 (1965).
The elements of the theory of algebraic groups can be found in
Borel, "Groupes linea ires algebriques", Ann. of Math. (2) 64, 20-82
(1956), or Mumford, "Abelian varieties", Oxford University Press,
London 1970.
So far there are no accounts of the general theory of schemes
having the character of a textbook. Mumford's mimeographed lecture
notes "Introduction to algebraic geometry", Harvard University Notes,
can serve as an excellent introduction, and a very full account is in
Grothendieck and Dieudonne's many-volume work "Elements de
geometrie algebriq ue". (Vol. I, Springer-Verlag, Berlin-Heidelberg-
New York 1971), which is not yet completed.
Advice to the Reader xv
In our book the number-theoretical aspect of algebraic geometry

is almost nowhere touched upon, although this aspect played a very
important role in the development of this branch of mathematics and
several of the most brilliant applications are connected with it. An
idea of this circle of problems can be obtained from Lang's "Diophantine
geometry", Interscience, New York-London 1962, and Cassel's paper
"Diophantine equations with special reference to elliptic curves",
J. London Math. Soc. 41,193-291 (1966).
One can become acquainted with the "analytic" direction in the
theory of algebraic varieties and the closely related theory of analytic
manifolds in the book by Weil "Introduction a l'etude des varietes
Klihlhiennes", Hermann et Cie., Paris 1958, and Chern's "Complex
manifolds", Univ. of Recife, 1959.
Finally, a great help in the understanding of algebraic geometry
is familiarity with the works of the classical, above all the Italian
geometers. Of the vast literature I only mention a few works that are
least specialized: F. Enriques and O. Chisini, "Lezioni sulla teoria
geometrica delle equazione e delle funzione algebriche, 3 vols., Bologna
1915-1924; G. Castelnuovo and F. Enriques, "Die algebraischen
Flachen vom Gesichtspunkte der birationalen Transformationen aus",
Enzykl. d. math. Wiss., III, 3; F. Severi "Vorlesungen tiber algebraische
Geometrie", Leipzig 1921; O. Zariski, "Algebraic surfaces", second ed.
Springer-Verlag, Berlin-Heidelberg-New York 1971 (the basic text of
the book contains an account of the classical papers, and the appendix
their translation into the language of present-day concepts).
Part I. Algebraic Varieties in a Projective Space
Chapter I. Fundamental Concepts
§ 1. Plane Algebraic Curves

The first chapter is concerned with a number of fundamental concepts
of algebraic geometry. In the first section we analyse some examples,
which prepare us for the introduction of these concepts.
1. Rational Curves. The curve given by the equation

y2 =x 2 +x 3 (1)
has the property that the coordinates of its points can be expressed as
rational functions of a parameter. To derive this expression we observe
that the line y = tx through the origin of coordinates intersects the
curve (1), apart from the origin, in a single point. For by substituting
y=tx in (1) we obtain X2(t 2 _X-1)=0. The root x=O corresponds
to the point 0 = (0,0). Apart from this we have one other root
x = t 2 - 1. From the equation of the line we find that y = t(t 2 - 1).
Thus, we have the required parameterization
x = t2 - 1 , y = t (t 2 - 1), (2)
and we have also clarified its geometric meaning: t is the slope of the
line passing through the points (x, y) and 0, and x and y corresponding
to t are the coordinates of the point of intersection, other than 0, of
the line y = tx and the curve (1). We can represent this parameterization
even more intuitively by drawing any line not passing through 0
(for example, the line with the equation x = 1) and associating with
a point P the point of intersection Q of the line OP with the chosen
line (projection of the curve from 0) (Fig. 1). Here the coordinate on
the chosen line plays the role of the parameter t. Both from this
geometric interpretation and from (2) it is clear that the parameter t is
uniquely determined (for x ¥= 0) by the point (x, y).
Now we give a general definition of plane algebraic curves for
which such a representation is possible. As a preliminary we introduce
some concepts. We fix a field k. Henceforth we mean by points the
points of the (x, y)-plane whose coordinates belong to k.
4 Chapter I. Fundamental Concepts
A plane algebraic curve is the set of all points whose coordinates

satisfy an equation
f(x,y)=O, (3)
where f (x, y) is a polynomial with coefficients in k.
In what follows we assume that k is algebraically closed. This is
due to the fact that otherwise an algebraic curve could have too few
points. For example, if we restrict ourselves to points with real
coordinates, we would have to say that the equations x 2 +2y2 =0
Fig. 1
and X4 + y4 = 0 determine one and the same "curve" - the ongm of

coordinates. By considering points with complex coordinates we obtain
two distinct curves. This does not mean that we exclude from the
discussion points whose coordinates lie in a field k that is not
algebraically closed. On the contrary, many problems lead to this
situation.
Here are some examples.
If k is the field of real numbers, we come to the "ordinary"
geometry of algebraic curves. If k is the field of rational numbers,
then the problem of finding all the points of (3), that is, all the
solutions in rational numbers of the corresponding equations, is one
of the basic problems in the theory of Diophantine equations.
If k is the field with p elements (p a prime number), then we have
the problem of solving the congruence f (x, y) == 0 (mod p).
§ 1. Plane Algebraic Curves 5
In all these cases the investigation of points whose coordinates lie

in the algebraic closure of the corresponding field k turns out to be
very important.
If the polynomialf(x, y) splits into two factors:f = gh, then the curve
defined by it is the union of the two curves defined by the equations
g(x, y) = 0 and h(x, y) = 0, respectively. If f is irreducible, then the curve
dermed by it is also called irreducible. Since every polynomial is a product
of irreducible factors, every plane curve is the union of finitely many
irreducible curves.
An irreducible plane algebraic curve X defined by the equation
f (x, y) = 0 is said to be rational if there exist two rational functions
cp(t) and lp(t), of which at least one is not constant, such that
f(cp(t), 11' (t)) =0 (4)
identically in t. Clearly, if t = to is a value of the parameter different
from the finite set of values for which the denominators of cp and 11'
vanish, then the point (cp(to), 11' (to)) lies on X. Later we shall show
that for a suitably chosen parametrization cp, 11' this correspondence
between values of the parameter t and points of a curve is one-to-one if
certain finite sets are excluded from the values of the parameter as
well as the points of the curve. Here the parameter t can be chosen as
a rational function X(x, y) of the coordinates x and y. If the coefficients
of the rational functions cp and 11' belong to a subfield ko of k and
to E ko, then the coordinates of the point (cp(to), 11' (to)) also belong to k o.
This circumstance points to one of the possible applications of the idea
of a rational curve. Suppose that the polynomial f (x, y) has rational
coefficients. If we know that the curve (3) is rational and that the
coefficients of cp and 11' lie in the field of rational numbers, then the
parametrization x = cp(t), Y = lp(t) gives us all the rational points of the
curve, with the possible exception of finitely many, when t ranges over all
rational numbers. For example, all the solutions of the indeterminate
equation (1) can be obtained from (2) when t ranges over all the
rational values.
Another application of rational curves is connected with the integral
calculus. We assume that the Eq. (3) of a rational curve determines
y as an algebraic function of x. Then any rational function g(x, y) is a
(compound) function of x. The rationality of the curve (3) implies the
following important fact: for every rational function g(x, y) the
indefinite integral
Jg(x, y) dx (5)
can be expressed in terms of elementary functions. For since the curve
(3) is rational, it admits a parametrization x = cp(t), Y = 11' (t), where
cp and 11' are rational functions. Substituting this expression in the
integral (5) we reduce it to the form .f g(cp(t), 1p(t)) cp'(t) dt, which is the
integral of a rational function. Such an integral is known to be
expressible in terms of elementary functions. Substituting the expression
t = x (x, y) of the parameter in terms of the coordinates, we obtain an
expression ofthe integral (5) by elementary functions of the coordinates.
We now give some examples of rational curves. Curves of order 1,
that is, straight lines, are, of course, rational.
Let us show that an irreducible curve of X of order 2 is rational. We
take a point (xo, Yo) on X and consider the line through (xo, Yo) with
the slope t. Its equation is of the form
y - Yo = t(x - xu) . (6)
Let us find the points of intersection of the curve with this line. It is
sufficient to substitute the expression for y obtained from (6) in the
equation of X. So we obtain an equation for x
f(x, Yo + t (x -xu)) =0, (7)
which is easily seen to be of degree 2. We know one of the roots of the
quadratic equation, namely x = xu, because the point (xo, Yo) is
assumed to lie on the curve. We denote by A the coefficient of x in
the equation obtained after dividing (7) by the coefficient of x 2 . Then
for the remaining root we have x + Xu = - A, x = - Xo - A. Since t
occurs in the coefficients of (7), A is a rational function of t. Substituting
this expression for x in (6) we obtain also for y an expression in the
form of a rational function of t. These expressions, as is clear from
the trend of the argument, satisfy the equation of the curve and hence
show that the curve is rational.
This parametrization has an obvious geometric meaning: the point
(x, y) is associated with the slope of the line joining it to the point (x o, Yo),
and the parameter t with the point of intersection of the curve with the
line passing through (xo, Yo) and having the slope t. This point is uniquely
determined, because we are concerned with an irreducible curve of
the second order. Just as we have done in connection with (1), we can
interpret this parametrization as the projection of X from the point
(xo, Yo) onto a line that does not pass through this point (Fig. 2).
Observe that in the construction of the parametrization we have
used the point (x o, Yu) on X. If the coefficients of the polynomial
f (x, y) and the coordinates x o, Yo of this point belong to some subfield
ko of k, then so do the coefficients of the functions giving the
parametrization. For example, we can find the general form of the
solution in rational numbers of a Diophantine equation of degree 2 if
we know at least one solution.
The problem of the existence of at least one solution is fairly

delicate. It is solved by the so-called Legendre theorem (see, for
example, [7J, Ch. I, §7.2).
Now we consider another application of our parametrization.
As we have seen, the equation y2 = ax 2 + bx + c of degree 2 dermes
a rational curve. From this it follows that, no matter what the rational
J
function g(x, y) is, the integral g(x, Vax 2 + bx + c )dx can be expressed
in terms of elementary functions. Our parametrization gives an
explicit form of a substitution that reduces this integral to one of a
Fig. 2
rational function. It is easy to check that we arrive in this way at

Euler's well known substitution.
Now we pass to curves of the third order. At the beginning of
this section we have already had an example of a rational curve of
order 3. We show next that there exist also non-rational curves of
order 3. An example is the curve with the equation x 3 + y3 = 1. This
follows from the more general result that the curve with the equation
x n +yn=1 (8)
is non-rational for n > 2, if n is not divisible by the characteristic
p ofk.
Suppose that the curve (8) is rational and that x = <p(t), y = lp(t)
is a parametrization of it. We write the rational functions <p and 11'
in the form
p(t)
<p(t) = r(t) ,
where p, q and r are polynomials, which we may assume to be relatively

prime. By hypothesis, the relation
(p(t»)n + (q(t»)n - (r(t»)n = 0 (9)
must hold identically. Differentiating it and dividing the result by n

(which is possible, because n is not divisible by the characteristic of k),
8 Chapter 1. Fundamental Concepts
we obtain the relation

p(t)n- 1 • p' (t) + q(t)n-1 . q'(t) - r(t)n-1 . r' (t) = 0 . (to)
We consider (9) and (to) as a system of linear equations in pn-l,
qn-l, - r n- 1, with the matrix
(p q r)
\p' q' r' .
Solving this system by the usual method we see that pn-l, qn-1,
and - r n- 1 are proportional to qr' - rq', rp' - pr', and pq' - qp',
respectively. Since p, q and r are relatively prime, it follows that
pn- 11(qr' _ rq'), q"- l l(rp' - pr'), r"- l l(Pq' - qp').
We denote the degrees of p, q, and r by a, b, and c, with a;;.. b;;.. c, say.

Then the first relation gives (n - 1) a.;;;; b + c - 1, which together with
b .;;;; a, c .;;;; a, n;;.. 3 leads to a contradiction.
These examples lead us to the following question. How can we
recognize whether a given plane algebraic curve is rational? As we shall
see later, this question is connected with rather subtle concepts of
algebraic geometry.
2. Connections with the Theory of Fields. We now show that

the question raised just above can be stated as a question in the
theory of fields. For this purpose we associate with every irreducible
plane algebraic curve a certain field, just as a field is associated with
every polynomial, namely the least extension in which the polynomial
has a root.
Let X be an irreducible curve, given by the equation f (x, y) = 0,
as in § 1.1 (3). We consider rational functions u(x, y) = p(x, y)/q(x, y)
(where p and q are polynomials with coefficients in k) such that q(x, y)
is not divisible by f(x, y). We say that such functions are dermed on X.
Two functions p(x,y)/q(x,y) and P1(X,y)/q1(X,y) defined on X are said
to be equal on X if the polynomial p(X,y)q1(X, y)-q(x, y)pdx, y) is
divisible by f (x, y). It is easy to verify that rational functions, considered
to within equality on X, form a field. This is called the field of
rational functions on X and is denoted by k(X).
Obviously all the elements of k(X) can be expressed as rational
functions of x and y. Here x and y are algebraically dependent: they are
connected by the relation f (x, y) = O. Starting out from this fact it is
easy to check that the transcendence degree of k(X) is 1.
If X is a line, for example y = 0, then every rational function
<p(x, y) is equal on X to the rational function <p(x,O) of the single
variable x, therefore the field of rational functions on a line is the same

as the field of rational functions of a single variable x: k(X) = k(x).
Suppose now that X is rational and has the parametrization
x = <p(t), Y = 1p(t). With any rational function u = p(x, y)/q(x, y) we
associate the rational function u(<p(t), 1p(t)) of t obtained by substituting
<p and 1p for x and y. First of all, let us verify that this substitution
makes sense, in other words, that the denominator q(<p(t), 1p(t)) gives
a function of t that is not identically zero. Suppose that q(<p(t), 1p(t)) =0.
We compare this equation with § 1.1 (4). By giving t various values in
k we see that the equations! (x, y) == 0, q(x, y) = 0 have infmitely many
solutions in common. (We must recall that the field is algebraically
closed, hence infinite.) But this is possible only when the polynomials
! and q have a common factor, a simple result of elimination theory
(see [33], §27).
Thus, our substitution gives a definite result for any function
u(x, y) that is defined on X. Furthermore, since <p and 1p satisfy § 1.1 (4),
two functions u and Ul that are equal on X give, after substitution,
identical rational functions of t. Thus, to every element of k(X) there
corresponds a definite element of k(t). This correspondence is, of
course, an isomorphism of k(X) with a subfield of k(t). This isomorphism
leaves the elements of k fixed.
At this place we make use of a theorem on rational functions, which
is called Liiroth's theorem. It states that every subfield of the field of
rational functions k(t) containing k is of the form k(g(t)), where g(t)
is some rational function, in other words, that the subfield consists of
all rational functions of g(t). If the function g(t) is not a constant, then
the correspondence !(u)---*!(g(t)) defines evidently an isomorphism of
the fields of rational functions k(u) and k(g(t)). Therefore Liiroth's
theorem can be put in the following form: a subfield of the field of
rational functions k(t) containing k, but different from k, is itself
isomorphic to a field of rational functions. Liiroth's theorem can be
proved by starting out from simple properties of extensions of fields
(see [33], § 63). Applying Liiroth's theorem to our situation we see
that if the curve X is rational, then the field k(X) is isomorphic to the
field of rational functions k(t). Suppose, conversely, that for some
curve X given by the equation! (x, y) = 0 the field k(X) is isomorphic
to the field of rational functions k(t), and that under this isomorphism
x and y correspond to <p (t) and 1p (t). Since the relation! (x, y) = 0
holds in k(X), it is preserved under the isomorphism and gives
!(<p(t), 1p(t)) =0, which means that the curve X is rational.
It is easy to see that any field K) k of transcendence degree 1
over k and generated by two elements x and y is isomorphic to k(X),
where X is an irreducible plane algebraic curve. For since the
transcendence degree of Kover k is 1, x and y must be connected by an

algebraic relation. If this is f (x, y) = 0, with an irreducible polynomial
f, then clearly X can be taken to be the algebraic curve defined by
this equation. Hence it follows that the question on rational curves
raised at the end of § 1.1 is equivalent to the following question in the
theory of fields: when is a field K) k of transcendence degree 1 over k
and generated over k by two elements isomorphic to the field of
rational functions k(t) of single variable? The condition that K should
be generated over k by two elements is rather unnatural from the
algebraic point of view. It would be more natural to consider
extensions generated by any finite number of elements. However, we
shall show later that this would not lead to a more general concept.
In conclusion we mention that the preceding arguments enable us to
solve the problem of the uniqueness of the parametrization of a
rational curve. Let X be a rational curve. By Liiroth's theorem the
field k(X) is isomorphic to the field of rational functions k(t). Suppose
that under this isomorphism x and y correspond to cp(t) and tp(t). Then
we obtain a parametrization of X:x=cp(t), y=tp(t). Let us show that
this parametrization has the following properties:
1) every point (xo, Yo) E X, with finitely many possible exceptions,
can be represented in the form Xo = cp(t o), Yo = tp(t o) for some to;
2) for all points, with finitely many possible exceptions, this
representation is unique.
Suppose that under the isomorphism k(X)--+k(t) the function
x(x, y) goes over into t. Then the inverse isomorphism k(t)--+k(X) is
given by the formula u(t)--+u(X(x, y»). Bearing in mind that the two
correspondences are inverses of one another, we arrive at the relations
x = cp(X(x, y»), y= tp (X (x, y»), ( 1)
t = X(cp(t), tp(t») . (2)
The first relation gives 1). For if X(x, y) = p(x, y)/q(x, y), and q(xo, Yo) #0
(there are only finitely many points (xo, Yo) E X for which g(xo, Yo) = 0
because the polynomials q(x, y) and f (x, y) are relatively prime), then
we can consider the value X(xo, Yo). Suppose that the point (xo, Yo) is
such that X(xo, Yo) is different from the roots of the denominators of
cp(t) and tp(t) (for the same reason there are only finitely many points
(xo, Yo) for which this is not so). Then the formula (1) gives the
required representation for the point (xo, Yo). Similarly, from (2) it
follows that the parameter value t if it exists, is uniquely determined
by (x o, Yo), with the possible excep'l:ion of the fmitely many points for
which q(xo, Yo) =0.
Observe that we have proved 1) and 2) not for any parametrization

of a rational curve, but for a specially constructed one. For an arbitrary
parametrization 2) need not be true: for example, the curve y2 = x 2 + x 3
of § 1.1 (1) has, apart from the parametrization given by § 1.1 (2), also
the parametrization x = t 4 - 1, y = t 2 (t 4 - 1), which is obtained from the
first by replacing t by t 2 • Obviously the parameter values t and - t
now correspond to one and the same point on the curye.
3. Birational Isomorphism of Curves. If a plane algebraic curve

is not rational, the coordinates of its points can nevertheless often be
expressed in terms of the coordinates of the points of another, possibly
simpler, curve. Consider, for example, a curve of the form
( 1)
where I (x) is a polynomial of even degree 2n, and set I (x) =g(x) (x -oc),
where g(x) is of degree 2n - 1. We divide both sides of (1) by (x - oc)2n
and set
y
x - oc = u- 1 , =V • (2)
(x - oc)"
Then g(x)/(x - oc)2n-1 = 11 (u), where 11 is of degree not greater than
2n - 1, and (1) takes the form
~=hM· ~
It is clear that, conversely, the coordinates of a point of (1) can be
expressed rationally in terms of coordinates of a point of (3). For it
follows from (2) that -1 (4)
X=U +oc, y=vu-n •
The transformations (2) and (4) are inverses of one another. Thus,
the curves (1) and (3) are transformed into each other, and this can
sometimes turn out to be useful, because the degree of 11 is smaller
by at least 1 than that off
Here we come across a new type of connection that may exist
between algebraic curves. The notion of a rational curve is included in
this more general concept: the formulae (1) and (2) at the end of § 1.2
can be interpreted as saying that the curve X is transformed into the
line s = 0 in the (s, t)-plane.
Here are the precise definitions. Let X and Y be irreducible curves
given by the equations I (x, y) = 0 and g(u, v) = 0, respectively. A
rational mapping of X into Y is dermed as a pair of rational functions
cp(x, y) and tp(x, y) defined on X such that the function g(cp(x,y), tp(x,y))
vanishes on X. It is easy to verify on the basis of the argument in
§ 1.2 that for all points (xo, Yo) E X, except finitely many, the values
cp(x o, Yo) and tp(xo, Yo) are defined and that (cp(xo, Yo), tp(xo, Yo)) E y.
Two curves X and Yare called birationally isomorphic if there exist

rational mappings of X into Yand of Y into X that are inverse to one
another. In other words, there must exist a mapping of X into Y given
by functions <p(x,y) and tp(x,y), and a mapping of Yinto X given by
functions ~(u, v) and 11(U, v), such that
~(<p(x,y), tp(X,y))=x,} X
on,
11 (<p(x,y) , tp(x,y)) =y
<p(~ (u, v), 11 (u, v)) =U '}
on Y.
tp(~ (u, v), 11 (u, v)) =v
For example, the curves (1) and (3) are birationally isomorphic,
and (2) and (4) are the corresponding mappings.
So we arrive at one of the central problems of algebraic geometry:
how to classify plane algebraic curves to within birational isomorphism.
Even today we can hardly claim that there is an exhaustive solution
of this problem. Nevertheless, a number of strong results pointing in
the direction of a solution will be proved later.
Clearly, from the algebraic point of view the simplest type of
algebraic curves are curves birationally isomorphic to a line, that is,
rational curves. Having constructed in § 1.2 an example of a non-
rational curve we have shown incidentally that the solution of the
problem raised above is not trivial: not all curves are birationally
equivalent to each other.
Let X and Y be two irreducible birationally isomorphic plane
algebraic curves and suppose that the mappings of one into the other
are given by the formulae
(u,v)=(<p(x,y), tp(x,y)), (x,y)=(~(u,v), I1(U,V)).
As in the investigation of rational curves, we can establish a connection

between the fields of rational functions k(X) and k(Y) on these curves.
For this purpose we associate with any rational function w(x, y)
defined on X the function w(~(u, v), I1(U, v)) on Y. It is easy to verify
that in this way we obtain a mapping of k(X) into k(y), which is
an isomorphism of these fields. Conversely, if the fields k(X) and
k(Y) are isomorphic, then to functions x, y E k(X) there must correspond
under this isomorphism functions ~(u, v), 11(U, V)E k(y), and to functions
u, v E k(Y) functions <p(x, y), 11' (x, y) E k(X), and again a trivial verifica-
tion shows that the pairs of functions <p, 11' and ~, 11 determine a
birational isomorphism of the curves X and Y. Thus, two curves are
birationally equivalent if and only if their fields of rational functions
are isomorphic.
We see that the problem of classifying algebraic curves to within

birational isomorphism is the geometrical aspect of the natural
algebraic problem of classifying (to within isomorphism) extensions
of k, of transcendence degree 1 and generated by finitely many
elements.
In this problem it is natural not to confme our attention to fields
of transcendence degree 1, but to consider fields of arbitrary fmite
transcendence degree. We shall see later that this wider statement of the
problem also has a geometrical interpretation. However, here we have
to go beyond the limits of the theory of algebraic curves and have to
consider algebraic varieties of arbitrary dimension.
Exercises
1. Calculate the area of the loop of the curve (1) in § 1.1, the so-called folium Cartesii.
2. Show that the lemniscate, defined by the equation (X2 + y2)2 = a 2(x2 - y2), is
rational. Hint: consider the points of intersection of the lemniscate with the pencil of
circles X2 + y2 = t(x _ y).
3. Show that the curve of order 3 given by the equation y2 = X3 + Ax + B is rational
if and only if the polynomial X3 + Ax + B has mUltiple roots (the characteristic of the
ground field is not equal to 2).
4. Find a rational parametrization of the circle X2 + y2 = 1.
5. Find the general form of the solution of the equation in Exercise 4 in rational
numbers, and derive from it the well-known formulae for Pythagorean numbers
(see, for example, [32] Ch. 1, Exercise 9a) that is, the solutions of the equation
x 2 + y2 = Z2 in integers.
6. Show that the field of trigonometric functions, that is, the field of all rational
functions of sin cp and cos cp, is isomorphic to a field of rational functions. In particular,
every trigonometric equation can be transformed into an algebraic equation.
7. Show that the curve given in polar coordinates by the equation r = sin 3cp,
and in Cartesian coordinates by (x 2 + y2)2 = y(3X2 - y2), is rational.
8. Given 2n numbers cxi(i=1, ... ,2n), CXi'iO, we set CXi+cxi-l=ai. Show that the
functions u = x + X-I, V = y/X' determine a mapping of the curve
y2= n2n (X-CXi) (X-CXi- l )

;=1
into the curve
v2 = n (u-a
2.
i=1
i)·
Is this mapping a birational isomorphism?

9. Prove that the curve given by the equation fn-I(X, y)+fn(x,y) =0 is rational if
it is irreducible. Here fn-I and fn denote homogeneous polynomials of degree n - 1
and n, respectively.
to. Show that the formulae u = x/( 1 - y), v = 0(1 + y)/( 1 - y) determine a rational
mapping of the curve X3 + y3 = 1 into the curve v2 = 4u 3 - 1. Is this mapping a birational
isomorphism? The characteristic of the ground field is not equal to 2 or 3.
11. Prove that every curve of the third order is birationally equivalent to the curve
in § 1.3 (1) with n = 2. Here the coefficients ofa parametrizing function belong to the field ko
if the curve contains a point x with coordinates in k o. Hint: draw the pencil of lines
through x and consider the pairs of points of intersection other than x of the lines of the
pencil with the curve. The characteristic of the ground field is not equal to 2.
12. Show that a curve of order 3 containing a point y with coordinates in ko is
birationally isomorphic to a curve § 1.3 (3) with n = 2 and that the coefficients of the
parametrization formulae belong to k o. Hint: draw the tangent at y and apply the
construction of Exercise 11 to its point of intersection, other than y, with the curve.
Show that the arguments at the beginning of § 1.3 are applicable to the curve so
obtained.
§ 2. Closed Subsets of Affine Spaces

Throughout what follows we are concerned with one and the same
algebraically closed field k, which we call the ground field.
1. Definition of Closed Subset. At various stages of the develop-

ment of algebraic geometry the idea of what its basic object is-"the
natural notion of an algebraic variety"-has changed. There have been
projective and quasiprojective varieties, abstract algebraic varieties,
schemes, algebraic spaces.
In this book algebraic geometry is treated in gradually increasing
generality. In the first chapters the most general concept, comprising
all the algebraic varieties to be studied there, is that of a quasi projective
variety. In the later chapters this role is played by schemes. We now
define one class of algebraic varieties, which plays a fundamental role
in all the subsequent definitions. Since the word variety is reserved for
the more general concepts, we use another term.
We denote by An the n-dimensional affine space over k. Its points
are therefore of the form ex = (ex 1, ... , ex,.), exi E k.
Definition. A closed subset in An is a subset Xc N consisting of all
common zeros of finitely many polynomials with coefficients in k.
Occasionally we speak briefly of a closed subset.
Henceforth we write a polynomial in n variables T1 , .•. , T,. in the form
F(T), understanding by T the set of variables T1 , ••• , Tn. If a closed
subset X consists of all the common zeros of the polynomials
F 1 (T), ... , F m(T) then we call F 1 (T) = ... = F m(T) = 0 the equations
ofX.
A set X defined by an infinite system of equations F ).(T) = 0 is
also closed. For, the ideal min the ring of polynomials in T1 , · · · , T,.
generated by all the polynomials F).(T) has a finite basis: m= (G 1 , ... , Gm).
It is easy to see that X is defined by the system of equations
G 1 = ... = Gm=O.
Hence it follows that the intersection of any number of closed sets
is closed. For if the X}. are closed, then to obtain a system of equations
§ 2. Closed Subsets of Affine Spaces 15
derming X =nx.
derming all the X ...
it is sufficient to take the union of the systems
The union of rmitely many closed sets is also closed. Clearly, it is

sufficient to verify this for the case of two sets. If X = XI U X 2, and if
X 1 is determined by the system of equations Fi(T) =0 (i = 1, ... , m)
and X 2 by the system Gj(T)=O (j= 1, ... , I), then X, as is easy to
verify, is determined by the system Fi(T) GiT) =O(i = 1, ... , m;j= 1, ... , I).
Let X be a closed subset of an affine space. A set U C X is called
open if its complement X - U is closed. Any open set U containing x
is called a neighbourhood of x. The intersection of all closed subsets
of X containing a given subset Me X is closed. It is called the closure
of M and is denoted by M. A subset M is called dense in X if M = X.
This means that M is not contained in any closed subset YCX, Y:;CX.
Example 1. The whole affine space JAn is closed: it is given by
the empty set of equations, or by the eqmition 0 = O.
Example 2. The subset Xc JAI consisting of all the points of JAI
other than zero is not closed: every polynomial F(T) that vanishes for
all T:;cO must vanish identically.
Example 3. Let us determine all the closed subsets Xc JAI . Such a
set is given by a system of equations F 1 (T) = 0, ... , F m(t) = 0 in a
single variable T. If all the Fi vanish identically, then X = JAI. If the
polynomials Fi(T) are relatively prime, then they have no' common
roots, and X has no points. But if all these polynomials have the greatest
common divisor D(T), then D(T)=(T-IX1) .. ·(T-IXJ, and X consists
of the finite set of points T=IX I , ... , T=IX n •
Example 4. Let us determine the closed subsets Xc JA2 • They are
given by a system of equations
( 1)
where now T=(TI , T2 ). If all the Fi vanish identically, then X =JA2 •

Suppose that this is not so. If the polynomials F I, ... , Fm have no
common divisor, then, as we have shown in § 1.2, the system (1) has
only a finite (possibly empty) set of solutions. Suppose, finally, that all
the polynomials Fi(T) have a greatest common divisor D(T). Then
Fi(T) = D(T) G;(T), where now the polynomials Gi(T) have no common
divisor. Evidently X = XI U X 2, where Xl is given by the system of
equations G1 = ... = Gm = 0, and X 2 by the single equation D = O. As
we have seen, X 1 is a finite set of points. The closed sets given in JA2 by
one equation are plane algebraic curves. Thus, a closed set X CJA2
either consists of a finite (possibly empty) set of points, or it is the
union of a plane algebraic curve and a finite set of points, or it is the

whole of 1A2.
Example 5. With a point IX E Jl:{ with the coordinates (IX 1 , ••• , IXr)
and a point {3EiN with the coordinates ({Jl' ... ,{3s) we associate the
point (IX,{3)EJl:{+-s with the coordinates (lXI' ... ,IX" {31' ... ,{3J Thus,
Jl:{+s is identified with the set of pairs (IX,{3), IXEJl:{, {3EiN. Let XCJl:{
imd Y C iN be closed sets. The set of pairs (x, y) E Jl:{ +', X E X, Y E -Y,
is called the product of X and Yand is denoted by X x Y. This is also
a closed set. For if X is given by the equations Fi(T) =0 and Yby the
equations G/U) =0, then X x Y is given in Jl:{+s by the equations
Fi(T) = 0, Gj(U) = O.
Example 6. A set X ClAn given by a single equation F(Tu ... ,7;,) =0
is called a hypersurface.
2. Regular Functions on a Closed Set. Let X be a closed set in

an affine space IN: , and let k be the ground field.
Definition. A function f given on X and taking values in k is called
regular if there exists a polynomial F(T) with coefficients in k such that
f = F(x) for all points x E X.
(x)
For a given functionfthe polynomial F is, in general, not uniquely
determined. For example, without changing f we can add to it any
polynomial that occurs in the system of equations for X.
The collection of regular functions on a given closed set X
forms a ring and even an algebra over k if the operations of addition,
multiplication, and multiplication of elements by k are defined as in
analysis, namely by means of the same operations on the values at
each point x E X. The ring so obtained is denoted by k[X] and is called
the coordinate ring of the closed set X.
We denote by k[T] the ring of polynomials with coefficients in k
in the variables T1 , .•• , 7;,. Obviously we can associate with every
polynomial FE k[T] a functionf E k[X], regarding F as a function on
the set of points X. Thus, we obtain a ring homomorphism of k[T]
onto k [X]. The kernel of this homomorphism consists of all polynomials
FE k[T] that vanish at all the points x E X. Like every kernel of a
homomorphism, this set is an ideal of the ring k[T]. It is called the
ideal of X and is denoted by '!Ix. Clearly
k[X] = k[T]/'!I x .
Thus, the ring k[X] is determined by the ideal '!Ix.
Example 1. If X is a point, then k [X] = k.
Example 2. If X =Jl:{', then '!Ix =0, and k[X] = k[T].
§ 2. Closed Subsets of Affine Spaces 17
Example 3. Suppose that Xc JA.2 is given by the equation T1 T2 = 1.

Then k[X] = k[T1, T 1- 1] consists of all rational functions of T1 of the
°
form G(T1)/Tl', where n;;;;, and G(T1) is a polynomia1.
Example 4. Let us show that k[X x Y] =k[X]®kk[Y] for any
closed set X and Y. We define a homomorphism
cp:k[X]®k k[YJ---+k[X x Y]
by the condition
It is clear that in this way we do, in fact, obtain a regular function on

the set X x Y and that cp is an epimorphism, because the functions
rxj and /3j (in the notation of § 2.1, Example 5) belong to its image and
generate the whole ring k[X x Y]. To prove that it is a monomorphism
it is sufficient to verify that if {};} are linearly independent over k in
k[X], and {g) in key], then cp(};®g) are linearly independent in
°
k[X x Y]. The equation
~>ij};(X) g}y) =
i,j
implies for any fixed y that l>ijg/y) =0, from which it follows
that cij =0.
Since the ring k [X] is a homomorphic image of the polynomial
ring k[T], the Hilbert basis theorem for ideals holds in it. The following
analogue of Hilbert's Nullstellensatz is also valid in it: if the functions
f E k[X] vanish at all the points x E X at which the functions
gu ···,gm vanish, thenj'E(gl, ... ,gm) for some r>O. For suppose that
fis given by a polynomial F(T), and gj by polynomials Gj(T), and that
Fj=O (j=1, ... ,0 are the equations of X. Then the polynomial F(T)
vanishes at all points rx E JAn at which the polynomials G 1 , ••. , Gm,
F 1, ... ,FI vanish. For since P/rx) =0, we have rxEX, and then F(rx) =0
by hypothesis. Applying Hilbert's theorem to the ring of polynomials
we see that F'E(G 1 , ••• ,Gm ,F 1 , ••• ,F1) and hence thatj'E(gl, ... ,gm)
in k[X].
How is the ideal m: x of a closed set X connected with the system
°
of equations F 1 = ... = F m= of this set? We know that F j E m: x by
definition of m: x , therefore (FI, ... , Fm) C m: x . However, we do not
always have (F 1, ... , F m) = m: x . For example, if Xc JA 1 is given by the
equation T2 = 0, that is, if it consists of the point T = 0, then m:x
consists of the polynomials without a constant term. Thus, m: x = (T), but
(Fu ... ,Fm)=(T2). However, we can always give the same set by a
system of equations G 1 = ... =G1=0 such that (G 1, ... ,G1)=m:X • To
see this it is sufficient to recall that every ideal in the ring k[T] has a
finite basis. Let Gl' ... , G1 be a basis of the ideal '!Ix, so that
'!Ix = (G 1 , ... , Gz). Then obviously the equations G1 = ... = Gz= de- °
termines the same set X and have the required property. Occasionally
it is convenient even to assume that a closed set is given by the infinite
system of equations F = 0, where F are all the polynomials of the
ideal '!Ix. For if (F I> ... , Fm) = '!Ix, then all these equations are con-
sequences of the equations F 1 = ... = Fm = 0.
Relations between closed sets are often reflected in their ideals.
For example, if X and Yare closed sets in the affine space JA.n, then
X) Y if and only if '!Ix e '!I y. Hence it follows that with every closed
set Y contained in X we can associate the ideal Oy of the ring k [X]
consisting of the images of the polynomials FE '!Iy under the homo-
morphism k[T]~k[X]. Conversely, every ideal 0 of k[X] determines
an ideal '!I in k [T]: '!I consists of all the inverse images of the
elements of 0 under the homomorphism k[T]~k[X]. It is clear that
'!I) '!Ix. The equations F = 0, where F are all the polynomials in '!I,
determine a closed set Yc X.
From Hilbert's Nullstellensatz it follows that Y is empty if and
only if Oy = k[X]. Otherwise the ideal Oye k[X] could be written as
the collection of all functions IE k [X] that vanish at all the points of Y.
In particular, every point x E X is a closed subset and hence
determines an ideal mx e k[X]. By definition this ideal is the kernel
of the homomorphism k [X] ~ k that associates with every function
IE k[X] its value at x. Since k[X]/mx is a field, the ideal mx is
maximal. Conversely, every maximal ideal me k[X] corresponds to
some point x E X. For it determines a closed subset Yc X. For every
point Y E Y we have m C my, and since m is a maximal ideaL we
see that m = my. If u E k[X], then the set of points x E X at which
°
u(x) = is closed. It is denoted by V(u) and is called a hypersuriace in X.
3. Regular Mappings. Let X e IN' and Ye JA.m be closed sets.

Definition. A mapping I : X ~ Y is called regular if there exist m
regular functionsJ~, ... ,fm on X such thatI(x) = (fl(X), ... ,fm(x)) for all
XEX.
Thus, any regular mapping I: X ~JA.m is given by m functions
II> . ··,fm E k[X]. To verify that we are concerned with a mapping
I: X ~ Y (Y being a closed subset of JA.m) it is clearly sufficient to
check that the functions II, ... ,!'n as· elements of the ring k [X]
satisfy the equations of Y.
Example 1. The notion of a regular function on X is the same as that
of a regular mapping of X into /AI.
§ 2. Closed Subsets of Mille Spaces 19
Example 2. A linear transformation is a regular mapping.

Example 3. The projection f(x, y) = x determines a regular mapping
of the curve given by the equation xy = 1 into JA.l .
Example 4. The preceding example can be generalized as follows:
let Xc E be a closed set and F a regular function on X. Consider
the set' X'CJA.n+l given by the equation Fi(Tl, ... ,T,,)=O, where
Fi=O are the equations of X in E and T,,+lF(Tl , ... , T,,)=1. The
projection IP(x l , ... ,X,,+1)=(X 1 , ... ,x,,) is a regular mapping IP:X'~X.
Example 5. The mapping f(t) = (t 2 , t 3 ) is a regular mapping of the
line JA.l into the curve given by the equation x 3 = y2.
Example 6. Here is an example of great importance in number
theory. We assume that the coefficients of the equation F;l:T) of the
closed set X C E belong to the prime field IFp of p elements.
As we said' in § 1.1, the points of X whose coordinates lie in Fp
correspond to solutions of the system of congruences Fi(T) =0 (mod p).
We consider the mapping IP of the space E, defined by the formulae
IP(oc l , ... , ocn) = (ocf, ... , oc~) .
This is obviously a regular mapping. It is important that IP carries
X into itself. For if oc E X, that is, Fi(OC) = 0, then by a property of
fields of characteristic p and the fact that Fi(T) E IFp[TJ we have
Fi(ocf, ... ,oc~)=(Fi(OC1, ... ,OCn})P=0. The mapping IP:X~X so obtained
is called a Frobenius mapping. Its significance lies in the fact that the
points of X whose coordinates are contained in IFp are characterized
among all the points of X as the fixed points of IP. For the equation
ocr = OCi has as its solutions precisely all the elements of IFp'
Let us clarify how a regular mapping acts on the ring of regular
functions on closed set. We begin with a remark that refers to arbitrary
sets and mappings. If f : X ~ Y is a mapping of a set X into a set Y,
then we can associate with every function u on Y (with values in an
arbitrary set Z) a function v on X as follows: v(x)=u(f(x»). Clearly
the mapping v:X ~Z determined by v is the product of the mappings
u: Y~Z and f:X ~ Y. We denote the function v by f*(u). Thus,j* is
a mapping of the functions on Y into the functions on X. Suppose
now thatfis a regular mapping X ~ Y. The mappingf* carries regular
functions on Y into regular functions on X. For if u is given by a
polynomial F(T1' ... , T,,) andfby polynomials F l , ... ,Fm, then v=f*(u)
is obtained simply by substituting Fi for T; in F, that is, it is given by
the polynomial F(F 1, ... , F m). Furthermore, regular mappings can be
characterized as mappings carrying regular functions into regular
functions. For suppose that a mapping of closed sets f :X ~ Y is such
that for any function u regular on Ythe function f*(u) is also regular.
Then this is so, in particular, for the functions ti defined by the
coordinates T;(i = 1, ... , m) on Y. Consequently, the functions f*(t i) are
regular on X. But this means that f is a regular mapping.
We have seen that if a mapping f is regular, then f* is the mapping
f*: k [Y] ~ k [X]. From the definition of this mapping it easily follows
thatf* is an algebra homomorphism of keY] into k[X]. Let us show
that, conversely, every algebra homomorphism <p :k[Y] ..... k[X] has the
form <p = J*, where J is a regular mapping of X into Y. Let t 1 , ... , tm
be coordinates in the space JA..m in which Y is contained, regarded as
functions on Y. Obviously tj E keY], and hence <p(t i ) E k[X]. We set
<p (t i) = Si and consider the mapping J given by the formulae
J(x) = (Sl(X), ... ,sm(x)). Of course, it is regular. We show thatJ(x)E Y.
For if Hc'U y , then H(tl' ... ,tm)=O in keY], and hence also <p(H)=O
on X. Let XEX. Then H(f(x)) = <p(H)(x) =0, and this means that
J(X)EY.
Definition. A regular mapping J :X ..... Y of closed sets is called an
isomorphism if it has an inverse, in other words, if there exists a regular
mapping g: Y..... X, such thatJg= 1 gJ= 1.
In this case the closed sets X and Yare also called isomorphic.
Obviously, an isomorphism is a one-to-one mapping.
From what we have said above it follows that ifJis an isomorphism,
then J* is an algebra isomorphism between k[X] alld keY]. It is easy
to check that the converse is also true, so that closed sets are
isomorphic if and only if their rings of regular functions are isomorphic
over k.
The facts just proved show that the correspondence X ..... k[X]
determines an equivalence of the category of closed subsets of affine
spaces, (and their regular mappings) and a certain subcategory of the
category of commutative algebras over k (and their homomorphisms).
What this category is, in other words, what algebras are of the form
k[X], is clarified in Exercises 1 and 2. (See also Theorem 5 in § 3.)
Example 7. The parabola given by the equation y = XC is isomorphic
to a line, and the mappings J (x, y) = x, g(t) = (t, t l ) determine an
isomorphism.
Example 8. The projection J (x, y) = x of the hyperbola xy = 1 into
the x-axis is not an isomorphism, because this mapping is not one-to-one:
there are no points (x, y) on the hyperbola for which f(x, y) =0.
See also Exercise 7.
Example 9. The mappingJ(t) = (t 2 , t 3 ) of a line onto the curve given
by the equation x 3 = y2 is easily seen to be one-to-one. However, it is
§ 2. Closed Subsets of Affme Spaces 21
not an isomorphism, because the inverse mapping is of the form

g(x, y) = y/x, and the function y/x is not regular at the origin of
coordinates (see Exercise 5).
Example 10. Let X and Y be closed subsets of JR. Consider
X x YC .lA.2r (Exercise 5 in § 1.1) and the linear subspace LI C .lA.2 r
given by the equations t1 =u 1, ... ,tr=u" the so-called diagonal. With
every point Z EX n Y we associate the point q>(z) = (z, z) E .lA.2 r , which
obviously belongs to (X x Y)nLl. It is easy to check that the mapping
q>:X n Y-(X x Y)nLl so obtained determines an isomorphism between
X n Yand (X x Y)nLl. Making use of this fact we can always reduce
the study of the intersection of two closed sets to that of the
intersection of another closed set with a linear subspace.
Later we shall be mainly interested in concepts and properties of
closed sets that are invariant under an isomorphism. The system of
equations by which the set is determined does not necessarily have
this property: sets given in distinct spaces JR by distinct systems of
equations may be isomorphic. Therefore it would be natural to look
for an invariant definition of a closed set, independent of its realization
in some affine space. Such a definition will be given in Ch. V in
connection with the notion of a scheme.
Now let us find out when the kernel of the homomorphism
f* : k [Y] - k [X] corresponding to a regular mapping f: X - Y is
trivial, in other words, when f* is an isomorphic embedding of keY]
in k[X]. Let us see when f*(u) =0 for uEk[Y]. This means that
u(f(x))=O for all points XEX. In other words, u vanishes on all
points of the image f(X) of X under the mapping f. The set of points
yE Yfor which u(y)=O is obviously closed; therefore, if it contains
f(X), then it also contains its closure f(X). Repeating the same
arguments in the reverse order we see that f*(u) =0 if and only if
u = 0 on f(X) or, what is the same, if u E 0f(X)' In particular, it follows
that the kernel of the homomorphism f* is 0 if and only if f(X) = Y,
that is, iff(X) is dense in Y.
This is necessarily so if f(X) = Y, but it can happen that f(X) =F Y,
yetf(X) = Y(see Example 3).
Exercises
va.
1. Let a be an ideal of a ring A. The set of elements a E A for each of which there exists
an integer n such that an. == 0 (mod a) is called the radical of a and is denoted by
Show that Va is also an ideal. Show that an ideal a E k [Tl' ... , T,.] is an ideal of a closed
subset of an affine space IAn if and only if Va = a.
2. Show that an algebra A over a field k is of the form k [X], where X is a closed set,
if and only if it is finitely generated over k and has no nilpotent elements (that is, from
an = 0, a E A, it follows that a = 0).
3. A set Xc /A2 is defined by the equations f: x 2 + y2 = 1 and g: x = 1. Find the

ideal ~x. Is ~x=(f,g)?
4. Let xc /A2 be the plane algebraic curve defined by the equation y2 = x 3 • Show
that all the elements of the ring k[X] can be written uniquely in the form P(x) + Q(x)y,
where P(x) and Q(x) are polynomials.
5. Let X be the curve of Exercise 4, and let f(t) = (t 2 , t 3 ) be a regular mapping
/AI ~x. Show that f is not an isomorphism. Hint: Use the result of Exercise 4 in
trying to construct a regular inverse mapping.
6. Let X be the curve defined by the equation y2 = x 2 + x 3 , and f the mapping
/AI ~ X defined by the formula f(t) = (t2 - 1, t(t 2 - 1)). Show that the corresponding
homomorphismf* maps the ring k[X] isomorphically onto the subring of the polynomial
ring k[t] consisting of the polynomials get) for which g(1)=g(-1).
7. Show that the hyperbola defined by the equation xy = 1 and the line /AI are not
isomorphic.
8. Findf(/A2) for the regular mappingf :/A2~/A2 given by theformulaf(x,y) = (x,xy).
Is this set open in Jk? Is it dense? Is it closed?
9. The same as in Exercise 8 for the mappingf :/A3~/A3 given by f(x,y,z) = (x,xy,xyz).
10. An isomorphism f: X ~ X of a closed set X into itself is called an automorphism.
Show that all the automorphisms of the line /Ai are of the form f(x) = ax + b, a", O.
11. Show that the mapping f(x, y) = (x, y + P(x)), where P(x) is an arbitrary
polynomial in x, is an automorphism of /A2. Show also that these automorphisms
form a group.
12. Show that if f(XI, ... ,Xn)=(PI(XI, ... ,Xn), ... ,Pn(XI, ... ,Xn)) is an automorphism
of /An, then the Jacobian loP'/oxjl E k. Denoting the value of this Jacobian by J(f), show
that the correspondence f~J(f) determines a homomorphism of the group of all
automorphisms of /An into the group of non-zero elements of k.
13. Suppose thilt X consists of two points. Show that the ring k[X] is isomorphic to
the direct sum of two copies of k.
14. Letf:X ~ Ybe a regular mapping. The subset TC X x Y consisting of the points
of the form (x,f(x)) is called the graph of f Show a) that T is a closed subset of Xx Y
and b) that Tis isomorphic to X.
15. The mapping py: X x Y~ Y defined by the formula py(x, y) = y, is called
a projection. Show that for Z C X and a regular mapping f: X ~ Y we have
f(Z)=py((Z x Y)nT), where T is the graph of f, and Z x Y consists of all points
(Z,y),ZEZ, yE Y.
16. Show that for any regular mapping f: X ~ Y there exists a regular mapping
g: X -> X x Y which is an isomorphism of X with a closed subset of X x Yand for which
f = py 9 (any regular mapping can be split into an embedding and a projection).
17. Show that if X = UUo is a covering of a closed set X by open sets, then there
exist finitely many sets Uo" ... , Uo" such that X = Uo,u ... U Uo,.
18. Show that the Frobenius mapping ({J is one-to-one. Is it an isomorphism if,
for example, X =/AI ?
§ 3. Rational Functions
1. Irreducible Sets. In § 1.1 we have come across the concept of an
irreducible plane algebraic curve. Now we state an analogous concept
in the general case.
Definition. A closed set X called reducible if there exist closed
subsets X1CX, XzCX, X1#X, X 2 #X, such that X=X1uX z.
Otherwise X is called irreducible.
Theorem 1. Every closed set is the union of finitely many irreducible

ones.
Proof Suppose that the theorem is false for a closed set X. Then X
is reducible: X=X 1UX'1, and the theorem is false for Xl or X'l,
say Xl. Then Xl is reducible, and again one of the closed sets whose
union it is must be reducible. So we construct an infinite sequence
of closed sets X)X 1 )X 2 ) ... , X:l-X 1 , X 1 :1-X 2' .... Let us show
that such a sequence can not exist. If it did, then for the corresponding
ideals we should have
~XC~Xl C~X2' ... , ~X:I-~Xl' ~Xl :I-~X2'····
But such a sequence cannot exist because in a polynomial ring every
ideal has a finite basis, hence every ascending sequence of ideals
breaks off. This proves the theorem.
If in a representation X = UXi we have Xi C Xj for i:l- j, then we can
discard Xi from this representation. Repeating this several times we
arrive at a representation X = UXi in which Xd: Xj for i:l- j. Such a
representation is called an incontractible (or irredundant) decomposition
of X into irreducible closed sets, and the Xi are called irreducible compo-
nents of X.
Theorem 2. The incontractible representation of a closed set is unique.
Let X = UXi = U lj be two incontractible representations. Then
, J
Xi = XinX = Xin(u lj) = U (X;n lj). Since Xi is irreducible by hy-
j •
pothesis, for some j we have X;n lj = Xi> that is, Xi C lj. Interchanging
the two decompositions we see that for j there exists an i' such that
ljCXi,. Consequently XiC ljCXi" and since the decompositions are
incontractible, i' = i and lj = X;. This proves the theorem.
Next we state the concept of irreducibility of a closed set X in
terms of the ring k[X]. If X is reducible, X=X 1UX 2 , then since
X) Xl' X:I- Xl' there exists a polynomial F 1 that vanishes on Xl'
but not identically on X, and a similar polynomial F2 for X 2. But
then Fl· F2 vanishes both on Xl -and on X 2, hence on X. The
corresponding regular functions fl' f2 E k[X] have the property that
fl :1-0, f2 :1-0, fl· f2 =0. In other words, fl and f2 are divisors of zero
in k[X]. Suppose, conversely, that the ring k[X] has divisors of zero:
fl· f2 =0, fl :1-0, f2 :1-0. We denote by Xl and X 2 the closed subsets
of X corresponding to the ideals (il) and (i2) of k[X]. In other words,
Xi consists of all those points x E X for which /;(x) = 0 (i = 1, 2).
Obviously Xi :l-X, because /;:1-0 on X, and X =X 1 uX 2 , because
fl· f2 =0 on X, hence at every point x E X either fl(X) =0 or
f2(X) = O. Thus, a closed set X is irreducible if and only if the ring
k[X] has no divisors of zero. This in turn is equivalent to the fact

that 2lx is a prime ideal.
Theorem 3. The product of irreducible closed sets is irreducible.
Suppose that X and Yare irreducible, but X x Y=Zl UZ2,
Zj ':/= X x Y (i = 1, 2). Then for every point x E X the closed set x x Y con-
sisting of the points (x, y), where y is an arbitrary point of Y, is isomorphic
to Y, hence irreducible. Since x x Y=(x x y)nZ1)u(x x Y)nZ2),
either x x Y C Z 1 or x x Y C Z 2. We consider the set Xl C X consisting
of those points x E X for which x x Yc Z 1, and we show that this set is
closed. For any point y E Y the set Xy of those points x E X for
which x x y E Zl is closed: it is characterized by the fact that
(X x y)nZ l = Xy x y, and the intersection of the closed sets X x y
and Z 1 is closed. Since Xl = n
yeY
X y' we see that Xl is also closed.
Similarly, the set X 2 consisting of the points x E X for which x x Yc Z2
is closed. So we see that Xl U X 2 = X, and since X is irreducible,
it follows that X1=X or X2=X. In the first case Xx Y=Zl' in the
second X x Y=Z2. This contradiction proves the theorem.
2. Rational Functions. Every commutative ring without divisors of

zero can be embedded in a field, its field of fractions.
Definition. If a closed set X is irreducible, then the field of fractions
of the ring k[X] is called the field of rational functions on X. It is
denoted by k(X).
Recalling the definition of the field of fractions we can say that k(X)
consists of those rational functions F(T)/G(T) for which G(T) ¢ 2lx,
and we take it that F/G=Ft/G 1 if FG 1 -F 1 GE21x. This means that
k(X) can also be constructed as follows. Consider the subring
(l)x C k(Tl' ... , T,,) consisting of those rational functions f = P/Q,
P, Q E k [T], for which Q ¢ 2lx. The functions f for which P E 2lx form
an ideal M x , and k(X) = (l)x/Mx-
In contrast to a regular function on a closed set, a rational function
does not always assume a defmite value at a point of this set; for
example, 1/x at 0 or x/y at (0, 0). Let us clarify when this happens.
Definition. A rational function q> E k(X) is called regular at a point

XEX if it can be written in the form q>=f/g,f,gEk[X], g(x) ,:/=0. In
this case the element f(x)/g(x) of k is called the value of q>(x) and is
denoted by q>(x). .
Theorem 4. A rational function q> that is regular at all points of a

closed set is a regular function of this set.
Let cP E k(X) be regular at all points x E X. This means that for.

every point x there exist elementsfx, gxEk[X], gx(x):;60, such that
cP = fJgx· Consider the ideal a generated by all the functions gx'
x E X. It has a finite base, so that there exist finitely many points
Xl,":'X N, such that a=(gxI, ... ,gxN)' The functions gx, cannot have a
common zero x E X, because then all the functions of the ideal a
would vanish at x, whereas gx(x):;6 O. From the analogue to Hilbert's
Nullstellensatz it follows that a = (1), in other words, that there exist
N
functions Ul , ••• ,UNEk[X] such that 2: Uigx, =
i=l
1. Multiplying both
sides of this equality by cP and using the fact that cP = fX/gXI' we fmd that
N
cP =.2: uJx"
1=1
that is, cP E k[X]. This proves the theorem.
The set of points at which a rational function cp on a closed set X
is regular is non-empty and open. The first assertion follows from the fact
that cp can be represented in the form cp=flg, wheref,gEk[X], g:;60.
This means that there exists a point x E X for which g (x) :;6 O. Obviously
cp is regular at this point. To prove the second assertion we consider all
representations cp = j;/gi' For every regular function gi the set l'i C X
consisting of those points x E X for which gi(X) =0 is obviously
closed, hence U i = X - l'i is open. The set of points U at which cp is
regular, by definition, is of the form U =UU i and is therefore open.
This open set is called the domain of definition of cp. For any fmite
system of rational functions cp 1, ... , CPm the set of points x E X at which
they are all regular is also open and non-empty. The first assertion
follows from the fact that the intersection of finitely many open sets is
open, and the second from the following useful property: the intersection
of finitely many non-empty open sets of an irreducible closed set is
non-empty. For let Ui=X - Yi, i= 1, ... ,m; nUi=0. Then Yi:;6X and
UYi = X. But the l'i are closed sets, and we arrive at a contradiction to
the fact that X is irreducible. Thus, any finite set of rational functions
can be equated with a non-empty open set. This remark is useful in view
of the fact that a rational function cp E k(X) is uniquely determined
when it is specified on some non-empty open subset U C X. For if
cp (x) = 0 at all x E U and cp:;6 0 on X, then by taking anyone
representation cp = fig, f, g E k[X], we find that X is the union of two
closed sets: X =X l uX 2 , Xl =X - U, and X 2 is determined by f=O.
This contradicts the fact that X is irreducible.
3. Rational Mappings. Let X C IN! be an irreducible closed set.

A rational mapping X --+ Am is given by an arbitrary collection of m
functions CPl'"'' CPm E k(X): Now we defme the concept of a rational
mapping cP: X --+ Y, where Yis a closed subset of Am.
Definition. A rational mapping cp:X -+ YeAm is a collection of m

functions CPl, ... ,CPmEk(X) such that (CP1(X); ... ,CPm(X))EY for every
point x E X at which all the functions CPi are regular. This mapping
cP is said to be regular at such a point x, and the point (cp 1 (x), ... , CPm(x))
is called the image of x and is denoted by cp(x).
The set of points of the form cp(x), where x ranges over those
points X at which cP is regular, is called the image of X and is denoted
by cp(X). Thus, a rational mapping is not a mapping of the whole set X
into Y, but it necessarily determines a mapping of some non-empty
open subset U e X into Y.
The study of functions and mappings that are not defined at all
points is an essential difference between algebraic geometry and other
branches of geometry, for example, topology.
As was shown at the end of the preceding subsection, all the
functions cP;, and hence the rational mapping cP = (cp l, ... , CPJ, are
defined on a certain non-empty open set U e X. Therefore rational
mappings can be regarded as mappings of open subsets; but it must be
borne in mind that distinct mappings may have distinct domains of
definition. The same applies, of course, to rational functions. To verify
that functions cP l, ... , CPm determine a rational mapping cP: X -+ Y we have
to check that the functions (CPl, ... , CPm), as elements of k(X), satisfy the
equations of the set Y. For if this property holds, then for any
polynomial u(Tl , ... , TJEm: y the function U(CPl' ... ,cpJ vanishes
on X. Therefore, at every point x where all the CPi are regular,
U(CP1(X), ... ,CPm(x))=O, that is, (CPl(X)' ... ,CPm(X))EY. Conversely, if we
have a mapping cP: X -+ Y, then for every u Em: y the function
U(CP1, ... , CPJ E k(X) vanishes on some non-empty open set U eX,
hence on X. It follows that U(CPl' ... , CPm) = 0 in k(X).
Let us clarify how a rational mapping acts on rational functions on a
closed set. We assume that for a rational mapping cp:X -+ Y the set
cp(X) is dense in Y. We regard cP as a mapping of sets U -+cp(X),
where U is the domain of definition of cP, and we construct its
corresponding mapping of functions. For every function J E k[Y]
the function cp*(f) is rational on X. For if YeAm and ifJis given by a
polynomial u(T1, ... , Tm), then cp*(f) is given by the rational function
U(CPl, ... ,CPm)· So we have a mapping cp*:k[Y]-+k(X), which is, of
course, a ring homomorphism of k[Y] into the field k(X). This
homomorphism is even an isomorphic embedding of k[Y] in k(X).
For if cp*(U) =0 for uEk[Y], this means that u=O on cp(X). But if
U # 0 on Y, then the equation U = 0 determines a closed subset
V(u) e Y, different from Y. Then cp(X) e V(u), and this contradicts the
fact that cp(X) is dense in Y. Clearly, the embedding cP* of k[Y] into
k(X) can be extended to an isomorphic embedding of the field of
fractions k(Y) of k[YJ into k(X). Thus, if cp(X) is dense in Y, then the
rational mapping cp determines an isomorphic embedding cp* of k(Y)
in k(X). If cp:X -+ Y and 11': Y-+Z are two mappings and if cp(X) is
dense in Y, then, as is easy to see, the product tpcp:X -+Z can be defined,
and if tp(Y) is dense in Z, then (tpcp) (X) is also dense in Z. For the embed-
dings of fields we then have the relation (tpcp)* = cp* 11'*.
Definition. A rational mapping cp: X -+ Y is called a birational
isomorphism if it has an inverse. This means that there exists a
rational mapping 11': Y-+X such that cp(X) is dense in Y, and tp(Y) in X,
and that tpcp = 1, cptp = 1. In that case X and Yare called birationally
isomorphic.
It is obvious that if a rational mapping cp: X -+ Y is a birational
isomorphism, then the embedding cp*: k( Y) -+ k(X) is an isomorphism.
It is easy to verify that the converse also holds (for plane algebraic
curves this was done in § 1). Thus, two closed sets X and Yare
birationally isomorphic if and only if the fields k(X) and k(Y) are
isomorphic over k.
Examples. In § 1 we have analysed a number of examples of birational
isomorphism between plane algebraic curves. Obviously, isomorphic
closed sets are bitationally isomorphic. In Examples 8 and 9 of § 2.3
the mappings, although not isomorphisms, are birational isomorphisms.
Closed sets that are birationally isomorphic to an affme space are
called rational. In § 1 we have come across rational algebraic curves.
Here are some other examples.
Example1. An irreducible hypersurface X determined in N by an
equation F(Tb ... , T,,) =0 of degree 2 is rational. The proof given in
§ 1.1 for n = 2 works in the general case. The corresponding mapping
can again be interpreted as a projection of X from some point x E X
onto a hyperplane I eN that does not pass through x. We only have
to take x so that it -is not a "vertex" on X, that is, (aFjaT;)(x) #0
for at least one i = 1, ... , n.
Example 2. Consider the hypersurface X in .A3 dermed by the cubic
equation x 3 + y3 + Z3 = 1 and suppose that the characteristic of the
ground field is not 3. There are some straight lines on X, for example,
the lines L1 and L2 given by the systems of equations
x+ y=O, x+ey=O,
L1: L2:
z=1, z=e,
where e is a cube root of unity, e # 1. The lines L1 and L2 are skew.
We describe a rational mapping of X onto a plane geometrically
and leave it to the reader to derive the formulae and to verify that we
are dealing with a birational isomorphism. We take a plane E in !A3

that does not contain L1 or L 2 . As is easy to verify, for x E X - L1 - L2
there exists a unique line L passing through x and intersecting L1
and L 2 . We denote the point of intersection LnE by f(x). This is the
required rational mapping X --+ E.
In algebraic geometry we are concerned with two equivalence
relations: isomorphism and birational isomorphism. Clearly, birational
isomorphism is a coarser relationship than isomorphism, in other
words, non-isomorphic closed sets may be birationally isomorphic.
Therefore often the classification of closed sets from the point of view
of birational isomorphism turns out to be simpler and more lucid than
from the point of view of isomorphism. Isomorphism, being defined
at all points, is close to such geometric concepts as homeomorphism
or diffeomorphism and therefore more convenient. An important
problem is the clarification of connections between these two equi-
valence relations. The point is: how much cruder is birational
isomorphism than isomorphism, in other words, how many closed
sets that are distinct from the point of view of isomorphism belong
to one and the same type from the point of view of birational
isomorphism? Later on we shall frequently come across this problem.
Both these equivalence relations can be defined purely algebraically:
closed sets X and Yare isomorphic if and only if the rings k[X]
and keY] are isomorphic, and they are birationally isomorphic if and
only if the fields k(X) and k(Y) are isomorphic. In this context it is
important to clarify what rings are of the form k[X] and what fields
of the form k(X), where X is an irreducible closed set. The answer is
very simple.
Theorem 5. An algebra A over afield k is isomorphic to a ring k[X],
where X is an irreducible closed set, if and only if A has no divisors of
zero and is finitely generated over k. An extension K of k is isomorphic
to a field k(X) if and only if it is finitely generated.
The necessity of all these conditions is obvious. If an algebra A
is generated by finitely many elements t 1, ... , tm then A ~ k[T1 , ... , T,,]/~,
where ~ is an ideal of the polynomial ring k [T1' ... , T,,]' Since A has no
divisor of 0, ~ is a prime ideal. Suppose that ~ = (F 1> ••• , F m).
Consider the closed set X C IN defined by the equations F 1 = ... = F m = 0;
we show that ~x=~, and then K[X] ~k[T1' ... , T,,]/~x~A.
If F E ~x, then by Hilbert's Nullstellensatz pr E ~ for some r >0.
Since ~ is a prime ideal, we then have F E~. Therefore 21x C 21, and
since the inclusion 21 C ~x is obvious, we have ~x =~.
If the field K is generated over k by the finitely many elements
t l' ... , tn, then the algebra A = k [t 1> .•• , til] satisfies the conditions of the
theorem, and by what we have already proved, A = k[X]. Since K is

the field of fractions of A, we have K = k(X).
In conclusion we prove one result that illustrates the concept of a
birational isomorphism.
Theorem 6. Every irreducible closed set X is birationally isonwrphic

to a hypersurface in some affine space IN.
Proof. The field k(X) is finitely generated over k, say, by the elements
t 1 , ••• , tno coordinates in IN regarded as functions on X.
Suppose that t 1 , ••• , t~ are algebraically independent over k, and that
d is the maximal number. Then every _element yE k(X) depends
algebraically on t 1 , ••• , t d , and there exists a relationshipf(t 1 , .•. , td, y)=O
for which the polynomial f (Tl' ... , 1'.J, I;, + 1) is irreducible over k.
Letf(T1 , .•• , 1'.J+l) be such a polynomial for t 1 , ••• , t H1 • We claim
that f'T,(T1 , ••• , 1'.J+ Ih~:O for at least one i = 1, ... , d + 1. For if this
were not the case, then all the 1; would occur in f with degrees that are
multiples of the characteristic p of k, that is, f would be of the form
f -- ..'t"' 'T'pit ···.Ld+l
ai,···id+,.Ll 'T'pid+' . We set ait···id+'-
- bPi,···id+"
g -_ 't"'b 'T'i 'T'id+'

.. it···id+' .Ll··· .Ld+l
and find that f = gP, a contradiction to the irreducibility off.

If fi,1=O, then the d elements t 1 , ••• ,ti - 1 , ti+l, ... ,td+1 are alge-
braically independent over k. For the element ti is algebraic over
the field k(t 1 , ••• ,ti - 1 , t i+1 , ••• ,td +1 ) because fT,1=O, and hence 1;
occurs in f Therefore, if the elements t 1 , ••• , t i- 1 , ti+ 1, ... , td +1 were
dependent, then the transcendance degree of the field k(tl' ... ,td +1 )
would be less than d, and this contradicts the independence of the
elements t 1 , ••• , td.
Thus, we can always renumber T1 , ••• , T" such that t 1 , ••• , td are
independent over k and that fi d + , 1= O. This shows that td + 1 is separable
over k(tl' ... , td). Since t d+2 is algebraic over this field, by Abel's
theorem on the primitive element we can find an element y such that
k(t 1 , ••• , td +2 ) = k(tl' ... , t d, y). Repeating the process of adjoining
elements t d+1, ... , tn we representthe field k(X) in the form k(z l' ... , Zd+ 1)'
where Z 1, ... , Zd are algebraically independent over k and
( 1)
the polynomial f is irreducible over k, and fi d+' 1= O. Obviously the

field of rational functions k(Y) over the closed set Y defined by (1) is
isomorphic to k(X). But this means that X and Yare birationally
isomorphic, and the theorem is proved.
Note 1. By virtue of the condition!{d+l#O in (1) the element Zd+l

is separable over k(z 1, ... , Zd)' Consequently, k(X)jk(z 1, ... , Zd) is a finite
separable extension.
Note 2. From the proof of Theorem 6 and Abel's theorem on the
primitive element it follows that Z 1, ... , Zd+ 1 can be chosen as linear
L1 CijX
/I
combinations of the original coordinates Xl' ""Xn:Zi= j

j=
(i=1, ... ,d+ 1). The mapping (Xl' ... ,Xn)-+(ZU ""Zd+l) given by these
formulae is a projection of An parallel to the linear subspace defined by
n
the equations L CijXj=O (i=1, ... ,d+1). This indicates the geometric
j= 1
meaning of the birational mapping whose existence is established in
Theorem 6.
Exercises
1. Let k be a field of characteristic "" 2. Decompose the closed set X C JA3 defined by
the equations x 2 + y2 + Z2 = 0, x 2 - y2 - Z2 + 1 = 0, into irreducible components.
2. Show that if X is the closed set of § 2, Exercise 4, then the elements of the field
k(X) have a unique representation in the form u(x) + v(x)y, where u(x) and v(x) are
arbitrary rational functions.
3. Show that the mapping of f of Exercises 5, 6, and 8 in § 2 are birational
isomorphisms.
4. Decompose the closed set X defined in JA3 by the equations y2 = XZ, Z2 = y3, into
irreducible components. Show that all its irreducible components are birationally
isomorphic to JAI .
5. Show that if a closed set X is defined in ffi:: by a single equation f.-I(TI , ... , 7;,)
+ fn(T.. ... , 7;,) =0, where f.-I and /., are homogeneous polynomials of degree n-l
and n, respectively, and X is irreducible, then it is birationally isomorphic to JAn - I.
(Such a closed set is called a monoid).
6. At what points of the circle given by the equation x 2 + y2 = 1 is the rational
function (1 - y)jx regular?
7. At what points of the curve X with the equation y2 =x 2 +X3 is the rational
function t = y/x regular? Show that t 11' k[X].
§ 4. Quasiprojective Varieties
1. Closed Subsets of a Projective Space. Let IPn be an n-dimensional

projective space, so that a point ~ E IPn is given by n + 1 elements
(~o : ... : ~n) of k and not all the ~i are O. Two points (~o: ... : ~n) and
(1'/0: ... : I'/n) are taken to be identical if and only if there exists a A. # 0
such that '1i = A.~i(i = 0, ... , n). For any collection (~o : ... : ~,,) determining
~ the ~i are called homogeneous coordinates of this point.
We say that a polynomial f(S) E k [So, '" ,SJ vanishes at a point
°
~ E]pn ifj(~o, ... , ~,,)= 0, no matter what coordinates ~i of ~ are chosen.
It is clear that then f(A.~o, ... , A.~,,) = for all A.:;6 0, A. E k. We write f in
the formf = fo + fl + ... + fr, where/; is the sum of all the terms of degree
i in! Then
f(A.~0, ... ,A.~")=fo(~o, ... , ~,,)+ A.fl(~O'·'" ~,,)+ ... + A.'f..(~o, ... ,~,,).
Since k is infinite, the equality f(A.~o, ... , A.~,,) = 0, which holds for all
A.:;6 0, A. E k, implies that/;(~o, ... ,~,,) = 0. Thus, if a polynomialjvanishes
at some point ~, then all its homogeneous components vanish at that
point.
Definition. A subset X C ]pn is called closed if it consists of all points

at which finitely many polynomials with coefficients in k vanish si-
multaneously.
The set of all polynomials f E k [So, ... ,SJ that vanish at all points
x E X form an ideal in the ring k[S], which is called the ideal of X and is
denoted by ~x. By what was said above, ~x has the property that if
a polynomial f is contained in it, then so are all its homogeneous com-
ponents. Ideals having this property are called homogeneous. Thus,
the ideal of a closed subset of a projective space is homogeneous. From
this it follows that it has a basis consisting of homogeneous polynomials:
it is sufficient to take any basis and to consider the system of all homo-
geneous components of the polynomials of the basis. In particular,
every closed subset of a projective space can be given by a system of
homogeneous equations.
Thus, to every closed subset XC]pn there corresponds a homogeneous
ideal ~x C k[So • ... , Sn]. Conversely, every homogeneous ideal ~ C k[S]
determines a closed subset X C lPn. For if F 1, ... , Fn is a homogeneous
basis of ~. then X is determined by the system of equations F 1 = 0, ... ,
Fn = 0. If this system has no solutions in the field other than the null
solution, then naturally X is taken to be given by the empty set.
In the case of closed subsets of affine spaces an ideal ~ C k[T]
determines the empty set only if ~ = (1): this is the content of Hilbert's
Nullstellensatz. In the case of closed subsets of a projective space this
need not be so: for example, clearly the empty set is also determined
by the ideal (So, ... , S,.). We denote by I. the ideal of k[S] consisting
of those polynomials in which only terms of degree at least s occur.
Obviously the ideal I. determines the empty set-it contains, for example,
the polynomials S~, which vanish simultaneously only at the origin.
Lemma. A homogeneous ideal me k[S] determines the empty set

if and only if it contains the ideal I.for some s > O.
We have already seen that Is determines the empty set. This is true
a fortiori for any ideal containing it. Suppose now that a homogeneous
ideal mc k[S] determines the empty set. Let F 1 , ••• , F m be a homo-
geneous basis of mand deg Fi = n. Then by hypothesis the polynomials
Fi(1, T l , ... , T,,), where 1j = SiSo· have no common roots. For a common
root (OCl, ••• , ocn) would give a common root (1, OCl' ••• , ocJ of F l' .••• , F m'
By Hilbert's theorem, there must then exist polynomials Gi(Tl , ••• , TJ
such that L F i ( 1, T l , ... , T,,) Gi(T1 , ••• , T,,) = 1. Substituting in these
i
equations 1j = SiSo and multiplying by the common denominator,
which is of the form S'O°, we find that S'O°E2L Similarly, for every i = 1, ... ,n
we can find an integer m i > 0 such that S'!'i E 21. If now m = max (mo, ... ,mJ
and s = (m - 1)(n + 1) + 1, then in every term S'O° ... S~n with
ao + ... + an ~ s at least one Si must occur with an exponent a i ~ m ~ m i ,
and since S'!'i E m, this term is contained in 21. This shows that Is C m.
Later we shall consider simultaneously closed subset of affine and
projective spaces. We call them affine and projective closed sets.
For projective closed sets the same terminology as for affine sets
is applicable, .namely if X and Yare two closed sets and Y C X, then
X - Y is called open in X. As before, the union of any number and the
intersection of a finite number of open sets are open, and the union
of a finite number and the intersection of any number of closed sets are
closed. The set.AO of points ~ = (~o: ... : ~J for which ~o #0 is obviously
open. Its points can be put into one-to-one correspondence with the
points of the n-dimensional affine space, by setting OCi = ~Jeo(i = 1, ... , n)
and assigning to the point ~ E AO the point (oc 1, ••• , OCn) E IA.n. Therefore
we call IA.~ an affine open subset. Similarly the sets 1A.'i(i = 0, ... , n)
consist of the points for which ~i # O. Clearly IPn = UlA.f
F or every projective closed set X C IPn the sets .U i = X nlA.7 are
open in X. As subsets of lA.i they are closed. For if X is given 'by the
system of homogeneous equations Fo= ... =Fm=O and degFi=ni'
then, for example, U0 is given by the system of equations
sc;njFj =Fit, T], ... , T,,) =0 U= 1, ... , m),
Ii = Si/SO(i = t, ... , n).
We call the Ui affine open subsets of X. Clearly X = U

Ui' A closed
subset U c.AO determines a closed projective set V, which is called
its closure and is the intersection of all projective closed sets containing
U. It is easy to check that homogeneous equations for V are obtained
by the process inverse to the one just described: if F (Tl' "', T,,) is any
polynomial of ~ and if deg F = I. then the equations of iJ are of the form

S~F(SdSo, ... , Sn/SO) = 0. Hence it follows that
U=iJn/A'Q. (1)
So far we have considered two objects that can lay claim to be called
algebraic varieties: affine and projective closed sets. It is natural to try
and find a single concept of which these two types of varieties would be
particular cases. This will be done more fully in Ch. V in connection with
the concept of a scheme. Here we introduce a more special concept,
which combines projective and affine closed sets.
Definition. A quasiprojective variety is an open subset of a closed
projective set.
Obviously a closed projective set is quasiprojective. For affine closed
sets this follows from (t).
A closed subset of a quasiprojective variety is defined as its intersection
with a closed set of a projective space. An open set and a neighbourhood
of a point are defined similarly. The notion of an irreducible variety and
the theorem on the decomposition of a variety into irreducible compo-
nents carries over verbatim from the case of affine closed sets.
A subvariety Yof a quasi projective variety X C ]pn is now defined as
any subset Y C X that is itself a quasiprojective variety in lPn. Obviously
this is equivalent to the fact that Y = Z - Z 1, where Z ) Z 1 and where
Z and Z1 are closed in X.
2. Regular Functions. Passing to the investigation of functions on

quasi projective varieties we begin with the projective space ]pn. Here
we come across an important difference between functions of homo-
geneous and of inhomogeneous coordinates: a rational function of
homogeneous coordinates
(1)
cannot be regarded as a function of a point x E]pn even when Q(x) # 0,

because the value f(a o, ... , an) changes when all the homogeneous
coordinates are multiplied by a common factor. However, homogeneous
functions of degree 0, that is, functions f = P/Q, where P and Q are
homogeneous of the same degree, can be regarded as functions of a
point.
If X is a quasiprojective variety, X C lPn, X E X, f = P/Q is a homo-
geneous function of degree 0, and Q(x) # 0, then f determines in some
neighbourhood of x a function with values in k. This function is called
regular in a neighbourhood of x or simply at x. A function given on X
and regular at all points x E X is called regular on X. All functions that

are regular on X form a ring, which is denoted by k[X].
Let us show that for a closed subset X of an affine space our definition
of a regular function is the same as that given in § 2. If X is irreducible,
then this is the content of Theorem 4 in § 3. In general, it is sufficient to
make insignificant modifications of the arguments by which this theorem
was proved. In it we understand regularity of a function in the sense of
the definition given in § 2.
By hypothesis, every point x E X has a neighbourhood U x in which
f = px/qx, where p.~ and qx are regular functions on X, and qx =1=0 on [fx.
Therefore
q.J = P., (2)
on u.~. But we may assume that (2) holds on the whole of X. To see this
it is sufficient to choose a regular function that vanishes on X - Ux but
not at x, and to multiply it by Px and q... Then (2) also holds outside U x'
because both sides of the equality vanish. As in the proof of Theorem 4
of § 3, we can find points Xl' .• ", X N and regular functions hI' ... , hN such
N
that L qXihi = 1. Multiplying (2) for x = Xi by hi and adding up we see
i=1
that N
f= L
;= 1
p..ih i ,
that is,! is a regular function.
In contrast to the case of affine closed sets, the ring k[X] can consist
of constants only. In §5 we shall show that this happens always when X
is a closed projective set. This can easily be verified directly when X = ]pn.
For if f = P/Q, where P and Q are forms of the same degree, we may
assume thatP and Q are relatively prime. Then the functionfis non-regul-
ar at points x where Q(x) = O. On the other hand, k[X] may tum out to
be unexpectedly large. Namely, if X is an affine closed set, then as a ring
k[X] is finitely generated over k. Rees and Nagata have constructed
examples of quasi projective varieties for which this is not so. This
shows that only for affine closed sets is the ring k[X] a natural invariant.
Now we pass to mappings. Every mapping of a quasiprojective
variety X into an affine space JAn is given by n functions on X with values
in k. If these functions are regular on X, then the mapping is called
regular.
Definition. Let f: X --+ Y be a mapping of quasi projective varieties
and Y C ]pm. This mapping is called regular if for every point x E X and
every open affine set JAr containing the point f(x) there exists a neigh-
bourhood U of x such that f(U) c JAr and the mapping f: U --+JAr is
regular. -
Let us verify that the property of regularity does not depend on the
particular open affine set IA,!, containing f(x) we have used. If
f,
f(x)=(yo, ... , ···,Ym)EIA7' is also contained in IAj, then Yj#O and
z · .
the coordinates of this point in IAj are of the form
(Yo/Yj' ... , 1/Yj' ... , f, ... , Ym/Y).
i j
Therefore, if the mapping f: U -+lAi is given by the functions

(fo, ... , f, ···,jm)' then f: U -+lAj is given by the functions
i .
(fo/ij, ... , 1/ij, ... , f, ·.. ,jm/ij)·

j
By hypothesis, ij(x) # 0 and the set U' of points of U at which ij # 0 is

open. On U' the functions f1/ ij, ... , l/fj' .. . ,fm/fj are regular, and hence
the mapping f: U' -+ IAj is regular.
As for affine closed sets, a regular mapping f: X -+ Y determines a
mappingf* : k[ Y] -+ k[ Xl
Now let us see by what formulae a regular mapping of an irreducible
variety is given in homogeneous coordinates. Suppose, for example,
that f(x) E IA'(J and that the mapping f: U --71A'(J is given by regular
functions fl, ... Jm· By definition,.r; = PJQ;, where Pi and Qi are forms
of the same degree in homogeneous coordinates of the point x and
Qi(X) # O. Taking these fractions to the least common denominator we
find that.r; = FJF 0, where all the F 0' ... , F m are forms of the same degree
and F o(x) # O. In other words. fix) = (F o(x): ... : F mix)), as a point in
IPm. In such a substitution we must bear in mind that the representation
of a regular function as a ratio of two forms is not unique. Therefore two
formulae
fix) = (F n(X): ... : F m(x)) ,
(3)
g(x) = (Go (x) : ... : Gm(x))
can give one and the same mapping. This is so if and only if
FiGj=FPi on X, O~i,j~m. (4)
So we arrive at a second version of the definition of a regular mapp-
ing.
A regular mapping f: X -+ IPm of an irreducible quasiprojective
variety is given by a collection of forms
(5)
of the same degree in homogeneous coordinates of a point x E IP". Two
mappings (3) are called identical if the conditions (4) hold. It is required
that for every point x E X there exists an expression (5) for f such that
F j (x):;6 0 for at least one i. Then the point (F o(x) : ... : Fm(x») is denoted
byf(x).
The importance of considering alI the expressions (5) of a regular
mapping is illustrated by the example of the projection of a conic onto
a line. If the curve is the circle xf + x~ = 1 and the centre of the projection
is the point (1,0), then the mapping is given by the formula t = X2/(X 1 - 1).
We introduce projective coordinates: Xl =ut!uo, X2=U2/UO, t=vt!vo.
Then the mapping can be written in the formf(uo:ul:U2)=(U1 -UO:U2)'
Both the forms U2 and Ul - Uo vanish at the point (1: 1: 0). But on the
circle u~ = (uo - ul)(UO+ u l ), and therefore the same mapping can be
given by the formula f(uo : Ut : U2) = ( - U2 : u 1 + u o). The form Ul + Uo
does not vanish at (1 : t : 0), which shows that f is regular.
Having defined a regular mapping of quasiprojective varieties, it is
natural to define an isomorphism as a regular mapping having a regular
inverse.
A quasiprojective variety X' isomorphic to a closed subset of an
affine space is called an affine variety. Here it can happen that X lies, but
is not closed, in /An. For example, the quasiprojective set X =/A 1 -0,
which is not closed in /A l, is isomorphic to a hyperbola, which is closed
in /A2 (Example 3 of §2.3). Thus. the concept of a closed affine set is not
invariant under isomorphism, whereas that of an affine variety is invariant
by definition.
Similarly, a quasi projective variety isomorphic to a closed projective
set is called projective variety. We shall show in §5 that if Xc IPn is
projective, then it is closed in lPn, so that the concept of a closed projective
set and a projective variety are the same and are invariant under iso-
morphism.
There exist quasiprojective varieties that are neither affine nor
projective (see Exercise 5 in §4 and Exercise 4,5, and 6 in §5).
Later we shall come across properties of a variety X that need only
be checked in an arbitrary neighbourhood U of any point X E X. In other
U
words, if X = U a' where the U~ are any open sets, then it is enough
to check such a property for each of the Ua' Such properties are called
local. Here is an example.
Lemma 1. The property of a subset Y C X of being closed in a quasi-

projective variety X is local.
This proposition means that if X = U U ex> where Ua is open and
Y II U a is closed in every U a , then Y is itself closed. By the definition of
open sets, Ua= X - Za' where Za is closed, and by the definition of
closed sets, Uall Y = Uall Ta, where the 4. are closed.
n
Let us verify that Y = (Za uTa), from which it follows, of course,
that Y is closed. If yE Y and yE Ua, then yE Ua. II Y C I'a, and if y¢ Ua,
then y E X - Ua = Za, so that y E ZI]. U 1'a for all 0(. Conversely, let
x E ZI]. U 1'a for all 0(. From the fact that X = U
U/l it follows that x E Up
for some p. Then x¢Zp, and hence XE Tp, XE Tpn Up C Y.
In studying local properties we confine our attention to affine
varieties.
Lemma 2. Every point x E X has a neighbourhood that is isomorphic
to an affine variety.
By hypothesis, X C lPn. If x E IA.'O (that is, the coordinate Uo of x is
not 0), then x E X n IA.'O, and by definition of a quasiprojective variety
X n IA.O = Y - Y1 , where Y and Yt C Yare closed subsets of IA.O. Since
x E 1, there exists a polynomial F in the coordinates in IA.O for which
F = 0 on Y1 , F(x) =1= O. We denote by (F) the set of points of the variety
Y where F = O. Obviously D(F) = Y - (F) is a neighbourhood of
x. We show that this neighbourhood is isomorphic to an affine variety.
Let F 1 = 0, ... , F m= 0 be the equations of Y in IA.'O. We define a variety
Z in IA.n+ 1 by the equations .
F 1 (T1 , ••• , T,,)= ... =Fm(Tl' ... , T,,)=O,
(6)
F(T1 , •••• T,,). T,,+ 1 = 1 .
The mapping <p: (x 1, ...• Xn + t> ---+ (x 1, ... , xJ clearly determines a

regular mapping of Z into D( F). and 1p: (x 1 , ••• , xn) ---+ (x 1 , .•. , Xn>
F(Xl' ... ,Xn)-I) is a regular mapping of D(F) into Z inverse of <po This
proves the lemma.
If Y = JA. 1, F = T, then Z is a hyperbola, and the isomorphism we have
constructed coincides with the mapping considered in Example 3
of §2.3.
Definition. An open set D(j) = X - V(f) consisting of points of an
affine variety X for which J(x) =1= 0 if E k[X]), is called a principal open
set.
The significance of these sets lies in the fact that, as we have seen,
they are affine and that it is easy to indicate their rings k[D(f)]. For by
construction,f =1= 0 on D(f), so thatF 1 E k[D(f)], and the Eqs. (6) show
that k[D(f)] = k[X] [11.fJ.
Lemmas 1 and 2 show, for example, that the images of closed subsets
under an isomorphism are closed. We can even show that under any
regular mapping J: X ---+ Y the inverse image J-l(Z) of any closed set
Z C Y is closed in X.
By definition of a regular mapping any points x E X and J(x) E Y
have neighbourhoods U of x and V ofJ(x) such thatJ(U) eVe IA.m and
thatJ: U ---+ V is a regular mapping. By Lemma 2 we may take U to be
an affine variety. By Lemma 1 it is sufficient to verify that J-l(Z) n U
=J- 1 (Z n V) n U is closed in U. Since Z n V is closed in V, it is defined
by equations gl = ... = gm = O. where gi are regular functions on V.

But then J- 1(Z n V) n U is determined by the equations f*(g 1) = ...
=f*(gJ = 0 and is therefore also closed.
From what we have shown it follows that the inverse image of an
open set is open.
It is easy to verify that a regular mapping can be defined as a mapping
J: X -+ Y such that the inverse image of every open set is open (continuity)
and that for every point x E X and every function <p that is regular in a
neighbourhood of the point J(x) E Y, the function f*(<p) is regular in a
neighbourhood of x.
3. Rational Functions. In defining rational functions on quasi-

projective varieties we are up against the fact that the general case is
entirely different from the case of affine varieties. Namely, for an affine
variety X we have defined rational functions on X as ratios of functions
that are regular on the whole of X. As we have seen, in the general case
it can happen that on a variety there are no functions other than constants
that are regular everywhere, and then there are no non-constant rational
functions. Therefore we define rational functions on a quasiprojective
variety X C lPn as functions that can be defined on X by homogeneous
functions on lPn. More accurately, we consider an irreducible quasi-
projective variety Xc lP n and denote (by analogy with §3.2) by (!Jx the
set of rational functions in homogeneous coordinates So, ... , Sn of the
formJ = P/Q, where P and Q are forms of the same degree and Q¢ m: x.
As for affine varieties, the irreducibility of X implies that (!Jx is a ring with-
out divisors of zero. We denote by M x the set of functions JE (!Jx for which
P E m:x. Clearly (!Jx/Mx is a field, which is called the field of rational
Junctions on the variety X and is denoted by k(X). Since a form vanishes
on an irreducible quasiprojective variety X precisely when it vanishes on
an open subset U, we have k(X) = k(U). In particular, k(X) = k(X), where
X is the closure of X in the projective space. Therefore, in studying fields
of rational functions we may restrict our attention, if we wish, to affine
or projective varieties.
It is easy to verify that if X is an affine variety, then the definition
above is the same as that in §3. For when we divide numerator and
denominator of a homogeneous function of degree zero J = P/Q, deg P
= deg Q = m, by S~, we can write it in the form of a rational function
of T; = S;/So(i = 1, ... , n). In this wayan isomorphism is established
between the field of homogeneous rational functions of degree zero in
So, ... , Sn and the field k( Tl , .... 1',.). A trivial verification shows that the
ring and the ideal of k(Tl' ...• Tn), which were denoted in §3.2 by (!Jx and
M x , then correspond to the objects we have here denoted by the same
letters.
Just as in the preceding subsection, where we have used the rational

functions on the space lPnto define regular functions, so we call a function
fE k(X) regular at a point x E X if it can be represented in the form
f = F /G, where F and G are homogeneous of the same degree, and
G(x) # O. Then f(x) = F(x)jG(x) is called its value at x. As in the case of
affine varieties, the set of points at which a given rational function f is
regular forms a non-empty open subset U of X. The set U is called the
domain of definition off Obviously rational functions can also be defined
as functions that are regular on open sets U C X.
A rational mapping f: X --+ lP m is determined (by analogy to the
second definition of a regular mapping given in §4.2) by specifying
m + 1 forms (F 0 : •.. : F J of n + 1 homogeneous coordinates of the
projective space lPn containing X. Here at least one of the forms must
not vanish on X. Two mappings (Fo: ... : FJ and (Go: ... : GJ are
called equal if F;G j = Fp; on X. When we divide all the forms F; by one
that is different from zero, we can give a rational mapping by m + 1
rational functions on X with the same concept of equality of mappings.
If a rational mappingf can be given by functions lfo: ... :fJ for which
all the fi are regular at x E X and not all vanish at this point, then the
mapping is regular at x. It also determines a regular mapping of some
neighbourhood of x into lPm.
The set of points at which a rational mapping is regular is open.
Therefore we can also define a rational mapping as a mapping of some
open set U C X. If Y C lPm is a quasi projective variety and f: X --+]pm
a rational mapping, then we say that f maps X into Y if there exists
an open set U C X in whichfis regular withf(U) C Y. The union U of all
such open sets is called the domain of regularity off, andf(U) the image
of X in Y.
As in the case of affine varieties, if the image of a rational mapping
f: X --+ Y is dense in Y, then it determines a field embeddingf* : k( Y) --+ k(X).
If a rational mapping f: X --+ Y has an inverse rational mapping, then
fis called a birational isomorphism, and X and Y birationally isomorphic.
In this case the embeddingf*: k(Y)--+k(X) is an isomorphism.
Now we can make the connection between the concepts of an iso-
morphism and a birational isomorphism more explicit.
Proposition. Two irreducible varieties X and Yare birationally iso-

morphic if and only if they contain isomorphic open subsets U eX and
VCY.
F or let f: X --+ Y be a birational isomorphism, let g: Y --+ X, 9 = f- 1 ,
be a rational mapping, and let U 1 C X and VI C Y be the domains of
regularity off and g. Since by hypothesisf(U 1) is dense in Y, we see that
f- 1 (V1) nUl is not empty and, as was shown in §4.2, is open. We set
V=f-I(V1)n VI' V=g-I(VI)n VI' A simple verification shows that

f(V) = V, g(V) = V,fg = 1, gf = I, that is, V and V are isomorphic.
4. Examples of Regular Mappings

1. Projection. Let E be a d-dimensional subspace of a projective
space IPn, determined by n - d linearly independent linear equations
LI = L z = ... = L n- d = 0, where Li are linear forms. The mapping
n(x) = (LI (x): ... : Ln-d(x)) is called a projection with centre at E. This
mapping is regular on lP n - E, because at the points of this set the forms
Li(i = 1, ... , n - d) do not vanish simultaneously. Therefore n(x) de-
termines a regular mapping n: X -7 IPn-d-l, where X is any closed
subset of IPn disjoint from E. The geometrical meaning of a projection
is the following. As a model oflP n - d - 1 we take any (n - d - i)-dimensional
subspace H C lP n disjoint from E. Through any point x E IP n - E and E
there passes a unique (d + I)-dimensional subspace Ex. This sub-
space intersects H in a unique point, namely n(x). If X intersects E but
is not contained in it, then the projection is a rational mapping. The
case d = 0, that is, a projection from a point, has already occurred
several times.
2. Veronese Mapping. Consider all homogeneous polynomials F of
degree m in the variables So, ... , Sn- They form a linear space whose
dimension is easily calculated to be (n!m).
We are interested in the varieties determined in IPn by an equation
F = 0. Such varieties are called projective hypersu~faces. Since propor-
tional polynomials determine one and the same hypersurface, the
hypersurfaces correspond to the points of a projective space lPvn,m of
dimension vn,m = (n!m) - 1. We denote homogeneous coordinates in
IPVn,m by via'" in' where i O ' ••• , in are any non-negative numbers such that
io + ... + in = m. Consider the mapping Vrn of lPn into IPVn.m defined by
the formulae
(1)
Obviously it is regular, because among the monomials on the right-hand
side of (1) there are, in particular, u,!" which vanish only when all the
U i = 0. This Vm is called a Veronese mapping, and vrn(IPn) a Veronese variety.
From (1) it follows that on vm(lP n)
(2)
if io + jo = ko + 10 , ... , ill + jn = kn + In· Conversely, from (2) it is easy to
°
derive that at least one coordinate of the form VO ... m ... o is different from
zero and that, for example, in the open set VmO ... O ¥ the mapping
U o = V mO ... O' U I = Vrn-I.l. O.... ,···, Un = Vrn-I,O ... I
is the inverse of Vm • Therefore vm(lPR) is determined by (2) and Vm is an iso-

morphic embedding of lP R into lP"n,m.
The significance of a Veronese mapping lies in the fact that if
F = L aio ... in U~ '" u~n is a form of degree m in the homogeneous co-
ordinates of the point x E IPR and if H is the hypersurface defined by the
equation F = 0 in IPR, then vm(H) is the intersection of vm(lPR) and the
hyperplane with the equation :Eaio ...in Vio ... in = 0 in IPvn,m. Therefore a
Veronese mapping makes it possible to reduce the study of certain
problems connected with hypersurfaces to the case of hyperplanes.
Exercises
1. Show that an affine variety U is irreducible if and only if its closure (j in a projective
space is irreducible.
2. Assign to any affine variety U contained in A:\ its closure (j in the projective
space lP·. Show that this gives a one-to-one correspondence between affine varieties in
A:\ and those projective varieties in lP' that have no component contained in the hyperplane
80 =0.
3. Show that the subvariety in lP 3 defined by the equations XIX 3 =x~, XOX2 =xt
XOX3 = XI X 2 (which is called a twisted cubic curve) is irreducible.
4. Decompose into irreducible components the variety defined by two of the equa-
tions in Exercise 3.
5. Show that the variety X = IA2 - X, where X = (0, 0), is not isomorphic to an affine
variety. Hint: Work out k[X] and use the fact that for an affine variety any proper ideal
21 C k[X] determines a non-empty subvariety.
6. Show that any quasiprojective variety is open in its closure in a projective space.
7. Show that any rational mapping cp:lPI ..... lP· is regular.
8. Show that any regular mapping cp : lP I ..... A' maps lP I into a single point.
9. Determine a birational isomorphismfbetween an irreducible quadric X in lP 3 and
a plane lP 2 , by analogy with Example I in §3.3 (stereographic projection). At what points
isfnot regular? At what point isf- ' not regular?
10. In Example 9, find open isomorphic sets U C X and V C lP 2 •
11. Show that the mapping Yo =X t X 2 , Yt =XOX2, Y2 = XOXI defines a birational iso-
morphism f of the plane lP 2 with itself. At what point is f and at what points is r 1 not
regular? Find open sets between which f determines an isomorphism.
12. Show that the variety vm(lP') is not contained in any linear subspace of lPvn,m.
13. Show that the variety of Exercise 3 coincides with V3(lP I ).
14. Show that the variety lP 2 - X, where X is a conic in lP 2 , is affine. Hint: Use a
Veronese mapping V 2 •
§ 5. Products and Mappings of Quasiprojective Varieties

1. Products. The definition of a product of affine varieties (Example 5 of
§2.1) was so natural that it did not require any explanations. For arbitrary
quasi projective varieties the matter is somewhat more complicated.
Therefore we begin by discussing quasi projective subvarieties of affine
spaces. If X C IAn and Y C lAm are such varieties, then the set X x Y
= {(x, y); x E X,y E Y} is a quasiprojective subvariety of IAn x IAm=lAn+m.
For if X=XI-X O' Y= YI - Yo, where XI' Xo and 11 , Y~ are Closed
subvarieties of the spaces IAn and lAm, respectively, then the representation
X x Y = X I X YI - (X I X Yo u Yj x X 0) shows that X x Y is quasi-
projective. We call this quasiprojective variety the direct product of X
and Y. At this place we have to verify that if X and Yare replaced by iso-
morphic varieties, then X x Y is also replaced by an isomorphic variety.
This is easy to check. Let <p: X ~ X' C lAP and lp: Y~ Y' C IAq be isomor-
phisms. Then (<p x lp):X x Y~X' x Y',where (<p x lp)(x, y)=(<p(x), lp(y)),
is a regular mapping, and (<p - \ lp - I) is its inverse.
We now turn to quasiprojective varieties and clarify what we expect
from the concept of a product. Let X C lP n and Y C lP m be two quasi-
projective varieties. We denote by X x Y the set of pairs (x, y), x E X,
Y E Y. We wish to regard this set as a quasiprojective variety, and for this
purpose we have to specify an embedding <p of it in a projective space
lPN such that <p(X x Y) is a quasi projective subvariety of lPN. Here it is
natural to ask that the definition should be local, in other words, that
arbitrary points x E X and y E Y should have affine neighbourhoods
U C X of x and V C Y of y such that <p(U x V) is open in <p(X x Y) and
that <p defines an isomorphism between the direct product of the affine
varieties U and V (whose definition is already known to us) and the
variety <pC U x V) C <p(X x Y). It is easy to see that by the property of
being local the embedding lPM is another such embedding, then lp<p - 1
determines an isomorphism between <p(X x Y) and lp(X x Y). For it is
sufficient to show that arbitrary x E X and y E Y have neighbourhoods
WI C<p(X x Y) of <p(x, y) and W2 C lp(X X Y) of lp(x, y) such that lp<p-l
defines an isomorphism between WI and Wz . With this aim we consider
affine neighbourhoods U C X of x and V C Y of y, whose existence is
guaranteed by the property of being local. We can even assume that
U x V is isomorphic to both <p(U x V) and lp(U x V), going over, if
necessary, to smaller affine neighbourhoods. Then <p(U x V) = f-l't and
lp(U x V) = Wz are the required affine neighbourhoods, because accord-
ing to the assumption we have made both are isomorphic to the direct
product U x V of the affine varieties U and V.
Let us proceed to the construction of the embedding <p having the
requisite property. Here we can at once confine ourselves to the case
when X = lPn, Y = lPm; if the embedding <p(lPn x lPm) ~ lPN has already
been constructed, then a simple verification shows that its restriction to
X x Y C lP n x lP m has all the necessary properties.
To construct <p we consider the space lP(n+ I)(m+ I) -I in which homo-
geneous coordinates Wj,; are numbered by double suffixes i and j
(i=O, ... ,n;j=O, ... ,m).·If x=(uo: ... :Un)E]pn, y=(vo: ... :V"JE]pm,
then we set
lP(x, y) = (wij), wij = u;v}i = 0, ... , n;j = 0, ... , m). (1)
Clearly, multiplication of the homogeneous coordinates of x (or y) by

a common non-zero factor does not change the point
lP(x, y) E ]p(n+ l)(m+ 1)-1 .
In order to show that lP(1Pn x ]pm) is a closed set in ]p(n +1)(m + 1) - 1, we
write down its equations,
WijWkl = WkjWiI (i, k = 0, ... , n;j, 1= 0, ... , m). (2)
Substitution shows that wij as defined in (1) satisfies (2). Conversely,
if the wij satisfy (2) and if woo ;6 0 say, then setting k = 1=0 in (2) we see
that (... : wij : ... ) = lP(x, y), where x = (woo: ... : WnO), Y = (Woo: ... wo"J·
This argument shows at the same time that lP(x, y) is uniquely determined
by x and y, so that IP is an embedding of]pn x ]pm in ]p(n+ 1)(m+ 1)-1. We
consider the open sets ~(uo;6 0) and ~(vo;6 0) in ]pn and ]pm, re-
spectively. It is clear that" .
o)
lP(iA!O x JA = Woo = JA~o+ 1)(m+ 1) - 1 1l1P(]pn X ]pm) ,
where JA~o+ 1)(m+ 1)-1 = {w oo ;6 OJ. If (W ij ) = IP(X, y) E Woo and zij = wiiwoo,
Xi = uJuo, Yj = vivo are inhomogeneous coordinates, then as we have
just found, ZiO =Xi, ZOj= Yj' zij=XiYj=ZiOZOj' for i>O, j>O. Hence it
follows that Woo is isomorphic to the affine space JAn+m with the co-
ordinates (Xl, ... , Xn' Yl' ... , Ym) and that IP determines an isomorphism
o o).
JA~ x JA -+ Woo = IP(~ x JA This proves that our construction has the
property of being local.
Note 1. The points (Wi) can be interpreted as (n+ 1)-by-(m+ 1)
matrices; the Eqs. (2) can be written in the form IWij Will
=0 and indicate
wkj Wkl
that the rank ofthe matrix (Wi) is 1, and the Eqs. (1) show that this matrix
is the product ofa column oftype (n + 1, 1) and a row of type (1, m + 1).
,
Note 2. The simplest case case n = m = 1 has a simple geometric
meaning. We have one Eq. (2) Wll woo = W 01 WlO so that 1P(]P1 x ]PI)
is a non-degenerate quadric Q in JP3. The set lP(rx x ]PI), where rx = (rxo : rxd,
is given in ]p3 by the equations rxlWOO=rxOWlO' rxlWOl=rxOWll and
determines a line in JP3. Similarly IP(JP 1 x {3), where {3 E]P\ is a line.
When rx ranges over the whole of ]p 1, the lines of the first kind give all
the lines of a family of generators of Q. The lines of the second kind give
the second family.
After having given the definition of a direct product by means of the
embedding IP of]pn x JPm in JP(n+ l)(m+ 1)-1, it is convenient to interpret
some concepts that were first defined by means of this embedding in

terms of the set IPn x IPm. For example, let us find out what subsets of
IPn x IP m go over under ({J into algebraic varieties. Subvarieties
X C IP(n+ 1)(rn+ 1) -1 are defined by equations FJw oo : ... : wnrn ) = 0, where
the Fi are homogeneous polynomials. After the substitution (1) this can
be written in the coordinates U i and Vj in theform Gi(U O : ... : un; Vo : ... : vJ
=0, where the Gi are homogeneous both in U o , "',Un and in V o , ... ,V rn ,
of the same degree of homogeneity with respect to the two systems of
variables. Conversely, as is easy to verify, polynomials with this property
of homogeneity can always be represented as polynomials in the products
uivj • However, if the equations are homogeneous both in the U i and vj '
then they always determine in IPn x IPm an algebraic subvariety even if
the degrees of homogeneity are different. If the polynomial
G( Uo : ••• Un; Vo : ... : Vrn)
is of degree r in the Ui and s in the Vj and if r > s, say, then the equation
G=O is equivalent to the system vi-sG=O(i=O, ... ,m) of which we
know already that it determines an algebraic variety.
Later we shall be faced with a similar problem for the product
o
IPn x JArn. Let JAm = JA C IPrn be given by the condition Vo 1= 0. The
equations of a closed set are of the form Gi(U O : ••• : Un; Vo : ... : Vrn) = 0.
Suppose that the Gi are homogeneous of degree ri in V o, ... , Vm • When
we divide the equations by v'& and set Yj = vivo, we obtain equations
gi(U O : ••. : Un; Y1, ... , Ym) = 0, where the gi are homogeneous in U o , ... , Um
but in general inhomogeneous in Yl ' ... , Yrn' So we have proved the
following result.
Theorem 1. A subset X C IPn x IPm is closed if and only if it is given by
a system of equations
°
Gi(U O : ••• : Un; Vo : ... : v".) = (i = 1, ... , t) ,
homogeneous in each system ()f variables U i and Vj separately. Every
closed subset ofIPn x JAm is given by a system ()f equations
gi(U O : ••• : un: Yl' ... , Ym) =O(i= 1, ... , t), (3)
homogeneous in the variables Uo' ... , Un'
Of course, the matter is similar for a product of any number of spaces.
F or example, a variety in IPn, x ... x lPn, is given by a system of equations
that are homogeneous in each of the I groups of variables.
2. Closure of the Image of a Projective Variety. The image of an
affine variety under a regular mapping need not be a closed set. For
mappings of an affine variety into another this is shown by Examples
3 and 4 of § 2.3. Fora mapping of an affine variety into a projective variety
this is even more obvious: an example is the embedding of JAn in IPn as an
open set INJ. In this respect projective varieties differ radically from
affine varieties.
Theorem 2. The image of a projective variety under a regular mapping
is closed.
The proof makes use of a concept which will also occur later. Let
f: X -+ Y be a regular mapping of arbitrary quasiprojective varieties.
The subset r f of X x Y consisting of the points of the form (x,J(x)) is
called the graph off
Lemma 1. The graph of a regular mapping is closed in X x Y.
First of all, it is sufficient to take Y as a projective space. For if
Y C ]pm, then X x Y C X x IPm,f determines a mapping J: X -+ ]pm, and
rf = r1 n (X x Y). Therefore we set Y = ]pm. Let i be the identity mapping
of ]pm onto itself. We consider the regular mapping if, i): X x ]pm -+ ]pm X ]pm,
if, i)(x, y) = (f(x), y)). Clearly r f is the inverse image of r; under the regular
mapping if, i). In §4.2 we have verified that the inverse image of a closed set
under a regular mapping is closed. Therefore everything reduces to a
verification that r; is closed in ]pm X ]pm. But r; consists of the points
(x, y) E]pm X ]pm, X = (uo : ... : um), Y = (vo : ... : vJ, for which (u o : ... : uJ
is proportional to (vo : ... : vrn). This can be written in the form U;Vj = UjV;,
wij = Wj; (i,j = 0, ... , m). The fact that ri is closed follows from Theorem 1,
and consequently the Lemma is proved.
We now return to the proof of the theorem. Let rf be the graph off and
p:X x Y-+ Ythe projection defined by p(x, y) = y. Obviously f(X) = p(rf ).
By Lemma 1, Theorem 2 is a consequence of the following more general
proposition.
Theorem 3. If X is a projective and Y a quasiprojective variety, then
the projection p : X x Y -+ Y carries closed sets into closed sets.
The proof of the theorem can be reduced to a very simple case. First
of all, if X is a closed subset of lPn, then by proving the theorem for ]pn
we also prove it for X because X x Y is closed in 1Pn x Y, and if Z is closed
in X x Y, then it is closed in ]pn x Y. Therefore we may assume that
X = ]pn. Secondly, since the concept of being closed is local, we may cover
Y by affine open sets U; and prove the theorem for each of them. There-
fore we may assume that Y is an affine variety. Finally, if Y is closed in
lAm, then ]pn x Y is closed in ]pn x lAm, so that it is sufficient for us to
prove the theorem when X = lPn, Y = lAm. What does the theorem mean
in this case? According to Theorem 1, every closed subset Z C ]pn X lAm
is given by a system of equations (3) of § 5.1, which we write in the form
g;(u; y) = O(i = 1, ... , t). Clearly if Yo E lAm, then p-l(yO) consists of all non-
°
zero solutions of the system g(u, Yo) = 0, and hence Yo E p(Z) if and only if
the system of equations g;(u, Yo) = has a non-zero solution. Thus,
Theorem 3 asserts that for any system (3) of § 5.1 the set T of those Yo E lAm
for which the system gi(U, Yo) = 0 has a non-zero solution is closed. By
Lemma 1 in §4.1 the system gi(U,YO)=O (i=I, ... ,t) has a non-zero
solution if and only if (gi(U, Yo), ... , gt(u, Yo») 1; Is for all s = 1,2, .... We
now verify that for any given s;;;, 1 the points Yo EArn for which
(gl (u, Yo), ... , gt(u, Yo») 1; Is form a closed set 1'.. Then T-=
closed. We denote by ki the degree of the homogeneous polynomial
n1'. is also
gi(U, y) in the variables uo, ... , Un' Let M(a) be an arbitrary numbering
of the monomials of degree s in the variables Uo,"" Un" The condition
(gl(U,yO)' ... , gt(U, Yo») ) Is means that all the M(a) can be represented
in the form t
Ma) = L gi(U, yo)Fi.a(u). (1)

i~ I
Comparing the homogeneous components of degree s we can see that

there must be a similar equation in which deg Fi,a = S - k i (and Fi,a = 0
if k i > s), We denote by N\P> the monomials of degree s - ki' numbered
arbitrarily. We see that the relation (1) is equivalent to the fact that all the
monomials M(a) are linear combinations of the polynomials gi(U, yo)N\P).
Of course, this is equivalent to the fact that the monomials g;(u, yo)N\P)
generate the whole linear vector space S of homogeneous polynomials
of degree s in Uo, ... , Un" Conversely, the condition
(gl (u, Yo), ... , gt(u, Yo»)JS Is
means that the polynomials gi(U, yo)N\P) together do not generate the
space S. In order to write down these conditions, we have to express the
coefficients of the polynomials gi(U, yo)N\Pl as a rectangular matrix and
equate to zero all the minors of this matrix of order (J = dimS. Clearly,
these minors are polynomials in the coefficients of the gi(U, Yo), hence
are polynomials in the coordinates of Yo' They give the equations of the
set Ts. Theorem 3, and hence also Theorem 2, are now proved.
Corollary 1. rr a function <p is regular on an irreducible projective
variety, then <p E k, that is, <p is a constant.
Proof We can regard <p as a mapping f:X ~Al, hence also as a
J
mapping l: X ~ lP l . Since <p is regular, so is f, and a fortiori Hence, by
Theorem 2 its image is closed. But sincefis regular andf(X)=l(X), the
set leX) is closed and is contained in Ai, that is, it does not contain the
point at infinity Xoo E }pI, From this it follows that either f(X) =Al or
that f(X) is a finite set S (Example 3 of § 2.1). The first case is impossible,
because f(X) must be closed in IP 1, whereas Al is not. Hence f(X) = s.
If S consists of the points ai' .". at' then X =U r
1 (a;), and if t > 1, this
contradicts the irreducibility of X. Therefore S consists of a single point,
which means that <p is constant.
Corollary 1 and Theorem 4 of § 3 are examples of the fact that affine
and projective varieties have diametrically opposite properties. On an
affine variety there is a wealth of regular functions: they form the whole
ring k[XJ, but the only regular functions on an irreducible variety are
constants. Here is another example of the contrast between affine and
projective varieties.
Corollary 2. A regular mapping f: X ~ Y of a projective irreducible
variety X into an affine variety Y maps X into a point.
Let Y C .lAm. The mapping f is given by m functions
f(x) = (CP1 (x), .•. , CPm(x»).
Each of.the functions CPi(X) is constant, by Corollary 1, CPi = IXi E k. There-
foref(X) = (IX1' ••• , IX;,,).
We give a further example of an application of Theorem 2. For this
purpose we make use of a representation of forms of degree m in n + 1
variables by points of the space 1P vn ,m (§ 4.4).
Proposition. Points eEIPVn,m corresponding to reducible polynomials F
form a closed set.
The proposition asserts that the condition of reducibility of a homo-
geneous polynomial can be written in the form of algebraic relations be-
tween its coefficients. For conics, that is, for the case m = n = 2, this relation
2
is well-known from analytic geometry: if F= L aijUiU j , then F is
i,j=O
reducible if and only if laijl = O.
Proceeding to the proof of the proposition we denote by X the set of
e
points E1P Vn ,m corresponding to reducible polynomials, and by
X h = 1, ... , m - 1) the set of all points corresponding to polynomials
X that split into factors of degree.i and m - j. Clearly X = X j' and we U
need only prove that each X j is closed.
Consider the projective spaces 1P Vn •i and IPvn.m-i. The multiplication
of two polynomials of degree j and m - j determines a mapping
f :IPVn,i x 1P Vn • m - j ~ IPVn.m,
which is regular, as is easy to see. Obviously Xj = f(IPVn.i X 1P Vn ,m- i ). As we

have seen in § 5.1, the product of projective spaces is a projective variety,
hence it follows from Theorem 2 that Xj is closed.
3. Finite Mappings. The projection mapping introduced in §4.4 has
an important property, which we can state after recalling some algebraic
concepts. Let B be a ring containing a ring A. An element b E B is said
to be integral over A if it satisfies an equation bm + a 1 bm - 1 + ... + am = 0,
ai E A. The ring B is called integral over A if each of its elements is integral
over A. It is easy to prove (see, for example, [37J, Vol. 1, Ch. V, § 1) that
a ring B having finitely many generators over A is integral over A if and
only if it is a module of finite type over A.
Let X and Y be affine varieties andf: X ~ Y a regular mapping such

thatf(X) is dense in Y. Thenf* determines an isomorphic embedding of
kEY] into k[X]. Making use of this we may regard kEY] as a subring
of k[X].
Definition 1. A mapping f is called finite if k[X] is integral over
k[Y].*
From the above property of integral rings it follows that the com-
positum of two finite mappings is itself finite. A typical instance of a
mapping that is not finite is Example 3 in §2.3.
Iff is a finite mapping, then every point y E Y has only finitely many
inverse images. For suppose that X C ffi.,n and that t 1, ... , tn are coordi-
nates in IA.n as functions on X. It is enough to prove that every coordinate
tj assumes only finitely many values on the set f-l(y). By definition tj
satisfies an equation tj + b 1tj-l + ... + bm= 0, b i E kEY]' For x EF l(y),
y E Y, we obtain the equation
(t/x))m+b1(y)(t}x))m-l+ ... +bm(y)=O, (1)
which has finitely many roots.
The meaning of the concept of finiteness lies in the fact that when y
ranges over Y, no root of (1) tends to infinity because the coefficient of the
leading term does not vanish. Therefore, as y ranges over Y, the points
.r-1 (y) can merge, but they cannot disappear. A more precise form of this
remark is the following theorem.
Theorem 4. A finite mapping is epimorphic.
Letf: X ~ Y be a finite mapping, X and Y affine varieties,y E Y. We
denote by my the ideal of kEY] consisting of the functions that vanish at
y. If t 1, ... , tn are coordinates as functions on Y and if y = (()(l, ... , ()(n),
then my = (t 1 - ()(1' •.. , tn - ()(n). The equations of the variety f-1(y) are
of the formf*(td = ()(l' ... ,1* (t,,) = ()(m and the setF l(y) is empty if and
only if,(f*(tl) - ()(1' ••. ,f*(tn) - ()(n) = k[X]. From now on we do not distin-
guish between functions uEk[Y] andf*(u) Ek[X], regarding kEY] as a
sub ring of k[X]. Then the preceding condition can be written in the form
(t1-()(1, ... ,tn -()(n)=k[X] or myk[X]=k[X]. From the fact that
k[X] is integral over kEY] it follows that k[X] is a module of finite type
over kEY]' In view of this, Theorem 4 follows from a purely algebraic
proposition.
Lemma 2. If a ring B is a module offinite type over a subring A (with
unit element), then aB #B for any proper ideal a C A.
Let B = Aw 1 + ... + AWn: then from aB = B it would follow
that Wi E aw 1 + ... + aw", that is, Wi = I ()(ijW j ' ()(ij E a. Therefore
° °
I (()(ij - bi)Wj = and hence dW j = (j = 1, ... , n), d = I()(ij - bi). Therefore
j
* Such a mapping is sometimes called finite and dominant.

dB = 0 and hence d = O. Since IXij E a, it follows from IIXij - c5ijl = 0 that

1 E a and hence a = A. This contradiction proves the lemma.
Corollary. A finite mapping carries closed sets into closed sets.
It is sufficient to verify this for irreducible closed sets. Iff: X ~ Y is
a finite mapping and Z C X is irreducible, then we have to apply Theorem 4
to the mappingl:Z~f(z), the restriction off to Z. It is obviously fmite,
hence f(Z) = f(z), that is, f(Z) is closed.
The property of being finite is local.
Theorem 5. Iff: X ~ Y is a regular mapping of affine varieties and if
every point x E Y has an affine neighbourhood U such that V = f- 1(U)
is affine and f: V ~ U finite, then f is also finite.
We set k[X] = B, kEY] = A. In §4.2 we have given the definition of a
principal open set. For every point we can take a neighbourhood U
being a principal open set and satisfying the conditions of the theorem
(see Exercise 11).
Let D(g,J be a system of such open sets, which we may take to be
U
finite in number. Then Y = D(g,,), that is, the ideal generated by all the
gaisequalto A. In our case, Va = f-l(D(ga)) = D(f*(ga)),k[D(ga)] = A [1/g a],
k[Va] =B[1/ga]. By hypothesis, B[1/ga] has a finite basis Wi,a over
A[1/ga]. Here we may assume that wi,a E B; if the basis consisted of
elements Wi,Jg':', wi,a E B, then the elements wi,a would also be a basis.
We consider the union of all the basis elements wi,a and show that they
form a basis of B over A.
Every element bE B has a representation
b= L ~w·
a.
nOl: L,a
i ga
for each IX. Since the elements g:« generate the unit ideal of A, there exist
ha E A such that L g:«ha = 1. Therefore
a
b = b L g:«h~ = L L a.,ahawi,a'
a I II
which proves the theorem.

Definition 2. A regular mappingf: X ~ Y of quasiprojective varieties
is called finite if every point y E Y has an affine neighbourhood V such
that the set U = f- 1 (V) is affine and the mapping of affine varieties
f: U ~ V is finite.
Obviously, for every finite mappingfthe setF l(y) is finite for every
yE Y.
From Theorem 4 it follows that every finite mapping is epimor-
phic.
This property leads to important consequences concerning arbitrary

mappings.
Theorem 6. Iff: X ~ Y is a regular mapping and f(X) is dense in Y, then
f(X) contains a set that is open in Y.
The assertion of the theorem easily reduces to the case when X and Y
are irreducible and affine, which we shall now assume. Then kEY] C k[X].
We denote the transcendence degree of the extension k(X)/k(Y)
by r, and we choose r elements U\, ... , U r E k[X] that are algebraically
independent over k(Y). Then k[X]) keY] [u l , ... , ur] ) kEY], and
k[Y][ul, ... ,urJ=k[YxAr]. Thus, the mappingfcan be represented
as the composition of two mappings: f = 9 h, h: X ~ Y x JA: and
0
9 : Y x Ar ~ Y, where 9 is simply the projection onto the first factor.

Every dement v E k[X] is algebraic over key x Ar], hence we can find
for it an element a E key x Ar] such that rJ. v is integral over key x Ar].
We choose such elements a l , ••• , am for some system of generators
VI' .•. , Vm of the ring k[X], and we set F = a l ... am' Since in the open set
D(F) C Y x Ar the functions ai are invertible, the functions v in
D(h*(F)) C X are integral, that is. the restriction
h: D(h*(F)) -~ D(F)
is finite. By Theorem 4, D(F) C h(X). It remains to show that g(D(F))

contains a set that is open in Y.
Let
'i U
1; = such that F(y, ,) ~ 0. Therefore g(D(F)) ) D(Fa)·
°
where the T(a), monomials in the variables T 1 , ••• , 7;., are coordinates in
Ar. For points Y E Y for which not all the Fa(Y) = there exist values
Theorem 6 shows how much simpler regular mappings of algebraic

varieties are than continuous or differentiable mappings.
The famous everywhere dense twisting of the torus, that is, the
mapping
gives an example of a situation that cannot occur for algebraic varieties

by virtue of Theorem 6.
Theorem 7. If X is closed in IP n and X C IPn - E, where E is a d-di-
mensional linea;' subspace, then the projection n: X ~ IPn-d-1 with
centre at E determines a finite mapping X ~ n(X).
Proof Let Yo, ... , Yn-d-I be homogeneous coordinates in IPn-d-1
and let n be given by the formulae Yj = L/x), x E X(j = 0, ... , n - d -1).
Obviously, U i =n- I (A7- d- l )nX is given by the condition Li(x)#O

and is an affine open subset of X. We show that n: Ui--+Ai- d- I n n(X)
is a finite mapping. Every function g E k[UJ is· of the form
g = Gi(X O , ••• , xn)/ L';', where Gi is a form of degree m. Consider the
mapping n l : X --+ IPn-d: Zj = Lj(x) U= 0, ... , n - d - 1), zn-d = GJx),
where Zo, ... , Zn-d are homogeneous coordinates in IPn-d. It is a regular
F I = ... = F s = °
mapping and its image nl(X) is closed in IPn-d, by Theorem 2. Let
be its equations. Since X C IPn - E, the forms
Li(i = 0, ... , n - d - 1) do not vanish simultaneously on X. This means
°
that the point 0= (0: ... : 1) is not contained in n 1(X), in other words,
that the equations Zo = ... = Zn-d-I = F 1= ... = Fs = have no solution
in IP n - d . By Lemma 1 of §4.1 it follows that
(Zo, ... , Zn-d-I' F I , ... , Fs)) Il
for some 1>0. In particular, Z~-d E(Zo, ... , Zn-d-I,FI , ... ,Fs)' This means
that n-d- I
Z;,-d= L zjHj+ L FjPj ,
j~ 0 j= I
where H j and Pj are polynomials. Denoting by H(q) the homogeneous

component of H of degree q, we deduce that
ifJ(zo, ... , Zn-d) =Z~-d- ~ziHy-l) =0 on nl(X), (2)
The homogeneous polynomial ifJ is of degree I and as a polynomial
in Zn -d it has the leading coefficient 1:
1-1
ifJ=Z~d- L Al-j(ZO,···,Zn-d-tlZ~-d· (3)

j~°
Substituting the transformation formulae for n l in (2) we find that
ifJ(L'(;, ... ,L'::-d-I,GJ=O on X. with (fJ of the form (3). Dividing this
relation by L';'z we obtain the required relation
I-I
gl+ I AI_}x'3. ... , 1, ... ,X::'-d-!lgj=O,
.i~ 0
where Xr = Yr/Yi are coordinates in An-d-I. This proves the theorem.
An application of a Veronese mapping makes it possible to generalize
this result substantially.
Theorem 8. Let F 0' ... , Fs be linearly independent forms of degree m on
IPn that do not vanish simultaneously on a closed variety Xc IPn. Then
<p(x) = (F o(x) : ... : Fs(x)) determines a finite mapping X --+ <p(X).
Let vm: IPn --+ IPV n •m be a Veronese mapping, and Li linear forms on
IPv n.mcorresponding to the forms F; on lPn. It is clear that then <p = n V m , 0
where n is the projection determined by the forms Lo, ... , Ls' Since Vm
is an isomorphism between X and vm(X), the theorem follows from
Theorem 7.
4. Normalization Theorem. We consider an irreducible projective

variety X C IPn other than the whole of IPn. Then there exists a point
x E lPn, X ¢: X, and the mapping q> projecting X from x is regular. The
variety q>(X) C IPn-l is projective, by Theorem 2, and the mapping
q>: X ~q>(X) is finite, by Theorem 7. If q>(X) # IPn-1, then we can apply
the same arguments to it. In the end we arrive at a mapping X ~ ]pm,
which is finite as the composition of finite mappings. The result we have
proved is called the normalization theorem:
Theorem 9. For every irreducible projective variety X there exists a
.finite mapping q>: X ~ lP m onto a projective space.
The analogous fact is true for affine varieties. To prove this we
consider an affine variety X C An. We assume that iN' is open in ]pn and
denote by g the closure of X in IPn. Let X # An. We choose a point
x E IPn - An, x¢: g, and we consider the projection q>: g ~ IPn-l from
this point. Here X is projected into "finite" points of IPn- 1, that is, into
points of the affine space An-l = IPn-l n An. We may continue this
process as long as X # jAn, and as a result we ohtain a projection q> : g ~]pm
for which q>(X) =Am. So we have proved:
Theorem 10. For every irreducible affine variety X there exists a
finite mapping q>: X ~ Am onto an affine space.
Theorems 9 and 10 allow us to reduce the study of certain (rather
crude) properties of projectives of affine varieties to the case of a projec-
tive and affine space. For m = 1 this was the point of view of Riemann,
who considered algebraic curves as covering of the Riemann sphere
(1P1 over the field of complex numbers).
Theorem 10 indicates that a ring A without divisors of zero that is
finitely generated over a field k is integral over a ring isomorphic to a
polynomial ring. This result can also be proved directly. It is easy to
derive Hilbert's Nullstellensatz from it (see [37], Vol. II, Ch. 7, § 3).
Exercises
1. Show that a variety cp(IP' x IPS) does not lie in any linear subspace ofjp('+ I)(s+ 1)-1
other than the whole of jp(d I )ls+ I) - I.
2. Consider the mapping of varieties jpl x jpl-->IPL PI(X,y)=x, P2(X, y)= y. Show
that pdX) = P2(X) = jpI for every closed irreducible subvariety xc jpl X jpl, provided
that X does not belong to one of the following types:
a) a point (xo, Yo) E jpl X jpl;
b) a set Xo x jpl, where Xo is a fixed point ofjpl;
c) a set jpI x Yo.
3. Verify directly Corollary 1 to Theorem 2 for the case X = jpn.
4. Let X =1A2 - x, where x is a point. Show that X is not isomorphic to an affine
nor to a projective variety (see Exercise 5 in §4).
5. The same as in Exercise 4. for jp2 - X.
§ 6. Dimension 53
6. The same as in Exercise 4 and 5 for lP 1 X JA1.

7. Is the mapping!: JAl .... X, where X is given by the equations y2 =x 3 ,J(t) = (t 2,t3 ),
finite? -
8. Let X be a hypersurface in lPr , L a line passing through the origin of coordinates,
and CflL the projection of X parallel to L onto an (r - 1)-dimensional subspace not containing
L. Denote by S the set of all those lines L for which CflL is not finite. Show that S is an alge-
braic variety. Find S when r = 2 and X is given by the equation xy = 1.
9. Show that the intersection of affine open subsets is affine. Hint: Use Example
to of §2.3.)
10. Show that forms of degree m =/'/ in n + 1 variables that are /-th powers of forms
correspond to points of some closed subset of lP'm .•.
11. Let!: X .... Y be a regular mapping of affine varieties. Show that the inverse image
of a principal affine open set is a principal affine open set.
§ 6. Dimension
1. Definition of Dimension. In §2 we have seen that the closed alge-
braic subvarieties X C A 2 are finite sets of points, plane algebraic
curves, or A2 itself. This division into three types corresponds to the
intuitive notion of dimension: we have varieties of dimension 0, 1 and 2.
Presently we shall give a definition of dimension for an arbitrary alge-
braic variety.
How can we arrive at such a definition? First of all, the dimension
of an n-dimensional projective or affine space must, of course, be n.
Secondly, if there is a finite mapping X -+ Y, then naturally we assume
that X and Y have the same dimension. Since by the normalization
theorems (Theorem 9 and 10 of § 5) any projective or affine variety X has
a finite mapping onto some space lPm or Am, this m is naturally taken as
the definition of the dimension ofthe variety. But here the question arises
whether this is a good definition: could there perhaps exist two finite
mappings f: X -+An and g: X -tAm, with m i= n? Suppose that X is
irreducible. Then from the finiteness of a mappingf: X -tAn it follows
that the field of rational functions k(X) is a finite extension of the field
f* k(An), which in turn is isomorphic to the field k(T1 , •.• , 1',.). Therefore
k(X) has the transcendence degree n, and this is a characterization of the
number n, independent of the choice of the finite mapping f: X -tAn.
In this way we have motivated to a certain extent the following definition
of dimension.
Definition. The dimension of a quasiprojective reducible variety X
is the transcendence degree of the field k(X).
The dimension of reducible varieties is the maximum of the di-
mensions of its irreducible components.
The dimension of a variety X is denoted by dim X.
If Y is a closed subvariety of X, then the number dim X - dim Y is
called the codimension of Y in X and is denoted by codim Y or codimx Y.
Observe that if X is an irreducible variety and U is open in X, then

k(U) = k(X), hence dim U = dim X.
Example 1. dim An = dim IP n = n, because k(An) is the same as the
field of rational functions in n variables. Since by definition the dimension
is invariant under a birational isomorphism, we see that An and Am
are not birationally isomorphic when n =1= m. "
Example 2. A plane irreducible algebraic curve is of dimension 1,
as we have seen in § 1.
Example 3. If X consists of a single point, then obviously dim X = 0,
hence the same is true when X is a finite set. Conversely, if dim X = 0,
then X is a finite set. It is enough to verify this for an irreducible affine
variety X. Let X C An and t l' ... , tn be coordinates in An as functions
on X, that is, as elements of k[X]. By hypothesis, ti is aigebraic over k,
hence assumes opJy finitely many values. From this it follows that X
is finite.
Example 4. Let us show that if X and Yare irreducible varieties, then
dim (X x Y)= dim X + dim Y.
It is sufficient to consider the case when X and Yare affine varieties,
X CAN, YCAM. Let dimX=n and dimY=m, and let t 1 , •.. ,tN
and it 1 , ... , UM "be coordinates in AN and AM, regarded as functions on
X and Y, respectively, with t 1, ... , tn algebraically independent in k(X),
and Ul' ... ' Um in k(Y). By definition k[X x Y] is generated by the
elements t 1, ... , tN' U 1, ... , U M , and by the assumptions we have made
all these elements are algebraically dependent on t 1 , ••• , tno Ul' ... , Urn. It
remains for us to show that these latter elements are independent.
Suppose that there exists a relation between them, F(T; U) = F(Tl' ... , T",
U l' ... , U m) = 0 on X x Y. Then for any point x E X we have
F(x; U 1, ... , Urn) = 0 on Y. Since U1 , ••• , Urn are independent on Y, each
coefficient ai(x) of the polynomial F(x, U) is o. This means that the
corresponding polynomial alI;, ... , T,,) = 0 on X. Now we use the
independence of the elements t 1 , ••• , tn on X to deduce that all the
ai (T1 , ••• , T,,) =0, so that F(T; U) = 0 identically.
One-dimensional and two-dimensional algebraic varieties are called
curves and surfaces, respectively.
Theorem 1. If Y C X, then dim Y';;;;; dimX. If X is irreducible, Y is

closed in X, and dim Y = dim X, then X = Y.
It is enough to prove the assertion for the case when X and Yare
affine and irreducible.
Let Y C X C AN and dim X = n. Then among the coordinates t 1, ... , tN
any n + 1 are algebraically dependent as elements of k [X], that is, they
§ 6. Dimension 55
are connected by a relation F (tit, ... , t in +.) = 0 on X. A fortiori this holds

on Y. But this means that the transcendence degree of k(Y) is not greater
than n, that is, dim Y ~ dim X.
Let dim Y = dim X = n. Then some n among the coordinates t 1, ••• , tN
are independent on Y, say t l' ... , tn. A fortiori they are independent on
X. Let U E k[ X], u:#: 0 on X. Then u depends algebraically on t 1, ... , tn
on X, that is, it satisfies a relation
(1)
onX.
We can choose the polynomial on the left-hand side of (1) to be
irreducible, and then a,(t 1, ... , t n) :#: 0 on Y. A fortiori the relation (1)
holds on Y. Suppose that u = 0 on Y. Then it follows from (1) that
a,(t 1 , ••. , t n ) =0 on Y, and since t\, ... , tn' by hypothesis, are independent
on X, we have a,(T1, ... , T,,) =0 on the whole of JAN. This contradicts
the fact that a,(t 1, ... ,tn):#:O on X Thus, ifu=O on Y, then u=o on X.
As we have seen, an irreducible plane algebraic curve is of dimen-
sion 1. A generalization of this fact is the following result.
Theorem 2. All irreducible components of a hypersurface in JAN or

lPN are of codimension 1. -
Proof It is sufficient to consider the case of a hypersurface in JAN.
Suppose that a variety X C JAN is given by the equation F(T) =0. -To
the decomposition F = Fl ... F into irreducible factors there corresponds
"
a representation X = Xl U ... U X", where Xi defined by the equation
Fi = o. Clearly it is enough to prove the theorem for the varieties Xi.
We show that they are irreducible. If Xi were reducible, then there
would exist polynomials G and H such that G H = 0 on X;. G ¥- 0, H ¥- 0
on Xi. By Hilbert's Nullstellensatz it follows that F;I(GH)' for some
1>0. Owing to the irreducibility of F, it follows that F;lG or F;lH, and
this contradicts the condition G:#: 0, H:#: 0 on Xi.
We assume that the variable TN actually occurs in Fi(T) and show
that the coordinates t l' ..• , t N _ \ are algebraically independent on Xi.
In fact, from a relation G(tl' ... ,tN-1)=0 on Xi it would follow that
Fi IG' for some I> 0, which is impossible, because G does not contain
TN. Thus, dim Xi ~ N -1, and since X ¥-JA N, it follows from Theorem 1
that dim Xi = N - 1. -
Theorem 3. Any variety X C JAN whose components all have codi-

mension 1 is a hypersUijace, and the ideal ~x is principal.
It is enough to consider the case when X is irreducible. Since X :#:JAN
(because dim X = N - 1), there exists a non-zero polynomial F that
vanishes on X. By the irreducibility of X some irreducible factor H of
F also vanishes on X. Then A'H ) X, and since we have seen in the proof
of Theorem 1 that A'H is irreducible, by Theorem 2 X =A'H. If GE ~x,
by Hilbert's theorem HI G1, and since H is irreducible, we see that G E (H),
that is, ~x = (H).
The following variant of Theorem 3 is proved similarly.
Theorem 3'. Any subvariety Xc IP N, X ... X IP N, whose components
all have codimension 1 is given by a single equation, homogeneous in each
of the I groups of variables.
Instead of the uniqueness of the decomposition of a polynomial
into irreducible factors we need only use the uniqueness of the decom-
position of a polynomial that is homogeneous in each group of variables
into factors of the same kind. This can be derived from the fact that if
F(x o , ... , x N" Yo, ... , YN2' ... , Uo , ... , UN) is homogeneous in each of the
groups of variables (xo, ... ,xN,), ... ,(uo, ... ,uN) and F=G·H, then
G and H have the same property.
2. Dimension of an Intersection with a Hypersurface. If we attempt
to investigate the varieties defined by more than one equation, then we
come at once to the problem of the dimension of the intersection of a
°
variety with a hypersurface. We investigate this problem first for pro-
jective varieties. If X is closed in IPN and F ¥- is a form on X, then we
denote by X F the closed subset of X defined by the condition F = 0.
F or any projective variety X C IPN we can find a form G( U 0, ... , UN)
of any preassigned degree m that does not vanish on any of the compo-
nents Xi. It is sufficient to choose in each component Xi of X a point
Xi E Xi and to find a linear form L that does not vanish at any of these
points. For G we can take a suitable power of L. Let X be closed in IPN
and assume that the form F does not vanish on any component of X. By
Theorem 1, dimX F< dimX. We set XF=X(l), and applying to it the
same arguments we find a form F 1, deg F 1 = deg F, that does not vanish
on any component of X(l). So we obtain a sequence of varieties Xii) and
of forms Fi(i = 0, ... ) such that
X=X(O»)X(1») ... ,X(i+1)=X~~,Fo=F. (1)
By Theorem 1, dim X(i + 1) < dim X(i). Hence when dim X = n, then x(n + 1)
is empty. In other words, the forms F 0 = F, F l' ... , Fn do not vanish
simultaneously on X.
Now let X be an irreducible variety. We consider the mappmg
cp : X -+ IP n given by:
(2)
This mapping satisfies the conditions of Theorem 8 in § 5, and by this
theorem the mapping X -+ cp(X) is finite. But if X -+ Y is a finite mapping,
§ 6. Dimension 57
then as we have seen, dim X = dim Y. Therefore dim IjJ(X) = dim X = n,

and since IjJ(X) is closed in IPR, by Theorem 2 of§5, we see that IjJ(X) =IPR,
by Theorem 1 of §6.1. Suppose now that dimX(l)=dimXF <n-1.
Then in the sequence (1) already X(R) is empty. In other words, the
forms F 0, ... , FR-1 do not vanish simultaneously on X. This means that
the point (0: 0 : ... 0: 1) is not contained in IjJ(X), but this contradicts
the fact that IjJ(X) = IPR. So we have proved the following result.
Theorem 4. If a form F does not vanish on an irreducible projective
variety X, then dim X F = dim X - 1.
Corollary 1. On a projective variety X there exist subvarieties of any
dimension s < dim X.
Corollary 2. (Inductive d~finition of dimension.) For an irreducible
projective variety X we have dim X = 1 + sup dim Y, where Y ranges
over all the proper subvarieties of x.
Corollary 3. The dimension of a projective variety X can be defined

as the largest number n for which there exists a chain of irreducible sub-
varieties Yo ) Y1 ) ••• ) Y", Yi =1= Yi + 1 .
Corollary 4. The dimension n of a projective variety X can be defined
as N - s - 1, where s is the maximum dimension of the linear subspaces
ofIP N that are disjoint from X.
Let E C IPN be a linear subspace and dimE = s. If s ~ N - n, then E can

be given by not more than n equations, and a subsequent application
of Theorem 4 shows that dim (X n E) ~ 0, hence X n E is not empty
(the dimension of the empty set is - 1 I). Setting m = 1 in the construction
of the sequence (1), we obtain n + 1 linear forms L o, ... , LR that do not
vanish simultaneously on X. If E is the subspace defined by them, then
dim E =N -n-1, and X n E is empty.
Corollary 5. The dimension of the set of zeros of r forms F 1, .•. , Fr on an
n-dimensional variety is not less than n - r.
This is proved by applying Theorem 4 in succession r - 1 times.
Note. Corollary 5 yields a fairly strong existence theorem: if r.;;; n,
then r forms have a common zero on an n-dimensional variety. For
example (with X = IPR), n homogeneous equations in n + 1 unknowns
have a non-zero solution. From this existence theorem we can deduce
a number of important results.
10. On IP2 any two curves intersect (since any curve is given by a
single homogeneous equation). On a quadric Q C IP3 there are non-
intersecting curves, for example, lines of one and the same family of
generators. Therefore lP 2 and Q are not isomorphic. Since they are

birationally isomorphic (Example 1 in §3.3), here we have an example
of birationally isomorphic, but not isomorphic, projective varieties.
This example will occur again later.
2°. Theorem 3 is false even for curves on a quadric Q. Not every
curve CeQ can be defined by equating a single form of lP 3 to zero.
In fact, supposing that the two non-intersecting curves which we have
found above on Q are given by the equations F 1 = 0 and F 2 = 0, we come
to a contradiction to Corollary 5 according to which the system F 1 = 0,
F 2 = 0, G = 0 (the equation of Q) has a solution.
Theorem. 5. Under the assumptions of Theorem 4 all the components of
the variety X F have the same dimension dim X - 1.
We consider the finite mapping cp: X -t lPn (n = dim X) that was
constructed in the proof of Theorem 4. Let Ai be open affine sets covering
lPn; then cp - 1 (!Ni) = U; are affine open sets in X, as we can easily see by
applying a Veronese mapping with m = deg F. Obviously it is enough
to show that all the components of the affine variety X F n Ui have the
dimension n - 1 for all i = 1, ... , n. From here on our arguments refer
to a fixed U i which we denote by U. Clearly X F n U = V(f), where
f = F/F;, so that it coincides on U with a set of zeros of the regular
function fE k[U]. We have constructed a finite mapping cp: U -tAn,
given by n regular functions fl , ... ,fn, with fl =f
To prove that all the components of V(f) have the dimension n - 1
it is sufficient to show that their dimensions are not less than n -1. We
show that the functions f2' ... ,fn are algebraically independent on each
of these components. Let P E k[T2 , ... , T,,]' To show that R = P(f2, ... ,fn)
does not vanish on any component of V (f), we need only show that
from a relation R· Q = 0 on V (n, Q E k[ U], it follows that Q = 0 on
V (f). For if V(f) = U(I) U ... U U(t) is an incontractible decomposition
into irreducible components and R = 0 on U(l), we can take for Q any
function that vanishes on U(2) u ... U U(t), but not on U(l). Then
R· Q = 0 on V (f), but Q # 0 on V (f). By Hilbert's Nullstellensatz our
proposition can be rephrased as follows: if fl (R . Q)l for some I> 0,
thenfl Qrn for some m > O.
Thus, Theorem 5 is a consequence of the following purely algebraic
fact.
Lemma. Let B = k[T1 , ••. , T,.], let A) B be a ring without divisors
of zero that is integral over B, x = T 1 , Y = P(T2' ... , T,.) # 0, U E A. If
x I(YU)1 in A for some I> 0, then x Iurn for some m > O.
The only property of the polynomials x and y that we make use of
is that they are relatively prime in the ring k[Tl' ... , T,.]' Observe that
we may replace yl by z and u l by v, and we need only show that if x and z
§ 6. Dimension 59
are relatively prime in k[T1 , ••. , 7;.], then from x Izv in A it follows that
x Ivrn in A for some m > 0. Thus, the lemma asserts that in a certain
sense the property of the polynomials z and x in B of being relatively
prime is preserved on transition to the ring A, which is integral over B.
We denote by K the field of fractions of B. If an element tEA is
integral over B, then it is clearly algebraic over K. We denote
by F(T) E K[T] the polynomial of least degree with leading coefficient
1 such that F(t) = 0, the so-called minimal polynomial of t. Division
°
with remainder shows that any polynomial G(T) E K[T] for which
G(t) = is divisible by F in K[T]. Hence we can conclude that an element
t is integral over B if and only if F[T] E B[T]. For if t is integral and
G(t) = 0, where G E B[T] has the leading coefficient 1, then G(T) =
F(T) . H(T) in K(T). But from the fact that in B the decomposition into
prime factors is unique (remember that B = k [Tl' .. " 7;.]) it follows that
then F(T) E B[T] and H(T) E B[T], a simple consequence of Gauss's
lemma.
Now it is easy to complete the proof of the lemma. Let zv = xw,
v, WE A, and let F(T) = Tl + bi T l - I + ... + bl be the minimal polynomial
of w. Since W is integral over B, we have b i E B. It is easy to see that the
minimal polynomial G(T) of v is of the form (xllzl)F((zlx)T). Therefore
G(T)=TI+(xbdz)T I - 1 + .. , + (x1b1lz l) ,
(3)
Since v is integral over B, we have xib;/Zi E B, and as z and x are relatively

prime, Zi Ibi' From (3) it then follows that x IVI, This proves the lemma
and with it Theorem 5.
Corollary 1. ~f X c lPN is a quasiprojective irreducible variety, F is
a form that does not vanish identically on X, and X F is not empty, then
each of its components has codimension 1.
Proof By definition, X is open in some closed subset X of lPN. Since
X is irreducible, so is X, and consequently dim X = dim X. By Theorem 5,
(X)F = U Yi, dim Yi = dim X - 1. But, as is easy to see, X F = (X)F n X;
hence it follows that X F = U (Yi n X), and Yi n X is either empty or
open in Yi, therefore dim (Yi n X) = dim X - 1.
Usually one meets the special case of this corollary when X cAn
is an affine variety. If An C lPn, An =A~, then X F = V(f), where f = Flil;,
m = deg F. Thus, X F .coincides with the set of zeros of some regular
functionfE k[X].
Corollary 2. If X C lPN is a quasiprojective irreducible n-dimensional
variety and Y is the set of zeros of m forms on X and is not empty, then
each of its components is of dimension not less than n - m,
The proof by induction on m is obvious. Again, in the case of an

affine variety we can speak of the set of zeros of m regular functions on
X.
If X is projective and n ~ m, then we can claim that Y is not empty.
Theorem 6. If X and Yare quasiprojective irreducible varieties in
lPN, dimX=n, dimY=m, N';;;;n+m, and Xn Y::f= 0, then
dimZ~n+m-N
for each component Z of X n Y.

Clearly the theorem is of local character, so that we need only prove
it for affine varieties. Let X, Y C JA.N. Then X n Y is isomorphic to
(X x Y) n A C JA.2N (Example 10 of § 2.3). The theorem now follows
from Corollary 2 to Theorem 5, because A is defined by N equations.
For projective varieties, as above, the set X n Y is not empty as long as
N .;;;; n + m. Theorem 6 can be stated in a more symmetrical form in
which it generalizes at once to an arbitrary number of subvarieties:
n Y;.;;;;
r r
codimx L codimx Y;. (4)
i= t i= 1
3. A Theorem of the Dimension of Fibres. If f: X -+ Y is a regular

mapping of quasiprojective varieties and y E Y, then the set f-l(y) is
called a fibre over the point y. Clearly, a fibre is a closed subvariety.
This terminology is justified by the fact that X is stratified into the
disjoint fibres of distinct points y Ef(X).
Theorem 7. Iff: X -+ Y is a regular mapping of irreducible varieties,
f(X) = Y, dim X = n, dim Y = m, then m.;;;; nand
1) dimf-l(y)~n-mfor every point yE Y;
2) in Y there exists a non-empty open set U such that dim; 1 (y) = n - m
for yE u.
Proof of 1). Clearly, this is a local property relative to y, so that we
need only prove it by taking for Y any open set U C Y containing y and
for X the variety f-l(U). Therefore we may assume that Y is an affine
variety. Let Y C JA.N. In the sequence (1) of §6.2 for Y we find that y(m)
is a finite set: y(m) = Y n Z, where Z is defined by m equations and
y E Z. We can choose U such that Z nUn Y = y, and we assume there-
fore that Z n Y = y. The subspace Z can be defined by m equations
g 1 = 0, ... , gm = O. Thus, the system of equations gl =0, ... , gm = 0 defines
the point y on Y. This means that on X the system of equations
!*(gl) =0, ... ,!*(gm) =0 defines the subvariety f-1(y). Now 1) follows
from Corollary 2 to Theorem 5 (the affine case).
Proof of 2). We may replace Y by an open affine subset W of it, and
X by an open affine subset VCr l(W). Since V is dense in f-1(W),
§ 6. Dimension 61
f(V) is dense in W. Therefore f determines an embeddingf* : k[W] -+ k[V].

We now assume that k[W] C k[V], and hence that k(W) C k(V). Let
k[W] = k[w l , ... , wM],k[V] =k[v l , ... , vN].Sincedim W =m,dim V=n,
the field k( V) has over k( W) the transcendence degree n - m. Suppose
that VI' ... , Vn - m are algebraically independent over k(W) and that
Vi (i = n - m + 1, ... , N) are connected with them by the relations
Fi(Vi;VI"",Vn_m;WI"",WM)=O. We denote by Vi the restriction
of Vi to f- I (y) n V. Then
k[f-I(y)nV]=k[vI"",VN]' (1)
We regard Fi as polynomials in Vi' VI' ... , Vn - m taking WI' ... , WM into
the coefficients, and we denote by 1'; the subvariety of W defined by the
vanishing of the leading coefficient of this polynomial. We set Yo = U 1';,
U = W - Yo. Clearly U is open and not empty. If y E U, then none of the
polynomials Fi(T;; TI , ... , T,,-m' wt(y), ... , wM(y») vanishes, that is, all
the Vi are algebraically dependent on VI' ... , Dn _ m' In conjunction with
(1) this shows that dimr I (y) ,;;;;;; n - m, and 2) now follows by virtue
of 1).
This proves the theorem. In the next subsection we shall see that 2)
need not hold for all y, in other words, that the dimension of a fibre
can actually jump.
Corollary. The sets Yi = {y E Y, dimr I (y) ~ I} are closed in Y.
By Theorem 7, Y..-m = Y, and there exists a closed subset Y' C Y,
Y' '" Y such that 1'; C Y' for I> n - m. If Zi are the irreducible components
of Y', then dim Zi < dim Y and we can apply induction on dim Y to
the mappingf-I(Zi)-+Zi'
Theorem 7 leads to a criterion for irreducibility of varieties which
is often useful.
Theorem 8. Iff: X -+ Y is a regular mapping of projective varieties,
f(X) = Y, and if Y and all the fibres r I (y) are irreducible and of the same
dimension, then X is irreducible.
We set dimr I(y) = n. Suppose that X is reducible and that X = UXi
is an incontractable decomposition into irreducible components. By
Theorem 2 of § 5 all the f(X i) are closed. Since Y = Uf(X J and Y is
irreducible we have f(X;) = Y for certain Xi' From Y we discard the
union of those closed sets f(X i) that are different from Y, and we denote
the remaining open set by Y'. We set.r-I(y') = X' and X' = U Xj, where
the Xj are open subsets of those Xj for whichf(Xj) = Y. Letij: Xj-+ Y'
be the restriction off, and mj the minimum of the numbers dim.fj I (y).
yeY'
By Theorem 7, this minimum is attained on some open set U C Y', and
since Uij-I(y) = r I(y) is irreducible and of dimension n, we see that
j
thatmaxmj=n and for a certain valuej=jo, dimfi~ I(y)=nfor YEU, hence
for all Y E Y. But then f-l(y) = Uf; l(y) for every y E Y, dimij-l(y)..;; n,
dimij~ l(y) = n, and from the irreducibility of f-l(y) if follows that
f-l(y)= f j; I (y). But this means that X jo = X' and consequently X jo = X.
A very special case of Theorem 8 is the irreducibility of the direct
product of irreducible varieties, which was proved in §3.
4. Lines on Surfaces. After the hard work spent on the proof of

the theorems on the dimensions of intersections, naturally we wish to
see some applications of these theorems. As an example, we now treat
the simple problem of the disposition of lines on surfaces in IP3.
As a rule, the concept of dimension turns out to be useful if we have
to give a rigorous meaning to the fact that an element of some set depends
on a given number of parameters. For this purpose we must identify the
set with some algebraic variety and apply our concept of dimension.
For example, we have seen that a hypersurface in IPn given by an
equation of degree m can be put into correspondence with points of the
projective space IPv n .... where vn•m= (n;;.m) -1 (see § 4.4).
We now turn to subvarieties that are not hypersurfaces; the simplest
of these are the lines in lP 3 •
A point of IP3 corresponds to a line of a four-dimensional vector
space E, and lines in IP3 to two-dimensional linear subspaces, that is,
planes in E. In an n-dimensional space E every plane FeE has a basis
of two vectors x, y which in turn determine a bivector co = x /\ YEA 2 E.
If the vectors x and y have in some basis the coordinates (Xl' ... , xJ
and (YI'"'' Yn), then the coordinates of the bivector x /\ yare
Pij=XiYj-YiXj (i,j=1, ... ,n). The bivector x/\y uniquely determines
a plane F, and under a change of basis x and yare multiplied by a non-zero
element of the field. Therefore the point of projective space with the
coordinates Pij is uniquely determined by, and uniquely determines, the
plane F.
In our case n = 4 and the bivector Pij has sixteen coordinates
Pij (i,j= 0, ... , 3), which are connected by the relations Pii=O, Pij= - Pji'
Thus, they are six independent coordinates POI' P02, P03' P12' P13, P23'
Every line L C IP3 determines a point (POI: P02: P03 : Pl2 : Pl3 : P23) E]PS;
Pij (0..;; i <j ..;; 3) are called the P1i.icker coordinates of L.
Not every point (Pij)(O..;; i..;;j..;; 3) corresponds to a line. For this it is
necessary and sufficient that the relation POIP23 - P02P13 + P03P12 =0
holds. Points satisfying this relation form a hypersurface of the second
order in ]ps, which we denote by JI. Clearly dim JI = 4. Thus, we have
set up a one-to-one correspondence between the lines L E]p3 and the
points of JI.
For the study of lines lying on surfaces the following result is very
important.
§ 6. Dimension 63
Lemma. The conditions for a line L with the PlUcker coordinates Pij
°
to belong to a surface X given by the equation F = are algebraic relations
between the Pij and the coefficients of F that are homogeneous both in
the Pii and in the coefficients of F.
We can write down a parametric representation of the coordinates of
the points of L in terms of its Plticker coordinates.
Let x and y be a basis of the plane AcE. Then the set of vectors
of the form
xf(y) - yf(x) (1)
coincides, as is easy to verify, with A as J ranges over all linear forms on

E. If the coordinates ofJhave the form (oco, OCI' OC2, O(3)' that is,/(x) = ~OCiXi'
then the vector (1) has the coordinates Zi = L. OCjPij' Pij = xiYj - YiXj'
j
Therefore a point of L having the Pli.icker coordinates Pij has the coordi-
nates U i = L. ocjPij' Substituting these expressions in the equation
j
F(u o, u l , U2' u 3 ) = °
and equating to zero the coefficients of all monomials
in oc i , we obtain the condition for LeX to hold in the form of algebraic
relations between the coefficients of F and the coordinates Pij'
Now we pass to the interesting question of the lines on surfaces on JP3.
F or a given m we consider the space r, v = v3 ,m = (m + 1)(m + 2)(m + 3)/6 - 1
whose points correspond one-to-one to surfaces of degree m in JP3
(that is, those given by a homogeneous equation of degree m). We
denote by Tm the subset of pairs (~, '1) E JP" x [J for which the line L
corresponding to the point '1 E [J is contained in the surface X corre-
sponding to the point ~ E JP". By the lemma Tm is a projective variety. Let
us find the dimension of Tm' For this purpose we consider the projections
cp(JPvxll)~JPv and 1p(IP"xll)~[J:cp(~,'1)=~, 1p(~,'1)='1. Obviously
cp and 1p are regular mappings. In what follows, we only consider them
on Tm. Observe that 1p(Tm) = rr. This means simply that through any
line there passes at least one surface of degree m, but perhaps a reducible
one.
We determine the dimension of the fibres 1p-I('1) of 1p. By making a
projective transformation we may assume that the line corresponding
to '1 is defined by the equations U o = 0, U I = 0. Points ~ E JPV such that
(~, '1) E 1p-l('1) C Tm , correspond to surfaces of degree m passing through
this line. The equation has the form F = 0, where F = Uo G + U I H, and
where G and H are arbitrary forms of degree m - 1. The set of these
forms corresponds, of course, to a linear subspace ofIP", whose dimension
is easy to compute. It is
11 = m(m + 1)(m + 5) _ 1. (2)
6
Thus,
d· -1() m(m+1)(m+5) 1
1m1p 1]= 6 -.
From Theorem 8 it follows that rm is irreducible. Applying Theorem 7

we see that ( 1)( 5)
· d· m m+ m+
dim rm = d1m 1p(rm) + 1m 1p- (1]) =
1
6 + 3 . (3)
We now consider the mapping <p: rm~lP\". According to Theorem 2
of §5, its image is a closed subset of lP\". Clearly dim <perm) ~ dim rm.
Therefore, if dim rm < v, then <perm) i= lPl", and this means that not every
surface of degree m contains a line. The inequality dim rm < v by (3)
means that m(m + 1)(m + 5)/6 + 3 < (m + O(m + 2)(m + 3)/6 - 1. It holds
when m> 3. So we have obtained the following result.
Theorem 9. For every m > 3 there exist sUl:faces of degree m that
contain no line. Furthermore, such sUl:faces correspond to an open set in
lP\".
Thus, there is a non-trivial algebraic relation between the coefficients
of a form F(u o, u 1 , u 2 , u 3 ) of degree m > 3 that are necessary and sufficient
for the surface given by the equation F = 0 to contain at least one line.
Of the remaining cases m = 1, 2, 3 the case m = 1 is trivial.
We consider the case m = 2, but the answer is well known from
analytic geometry.
From m = 2, v = 9, dim rm = 10. From Theorem 7 it follows that
dim <p - 1 (~) ;;. 1. This is a standard fact: every quadric contains infinitely
many lines.
Without going into the details of the proofs we note that here we come
across the phenomenon of the dimension of fibres jumping of which we
spoke in § 6.3. For if a quadric is irreducible, then for the points corre-
sponding to it dim <p - 1 (~) = 1, but if it degenerates into two planes, then,
of course, dim <p - 1 (~) = 2.
We now consider the case m = 3. Then dim rm = v = 19. It is easy to
construct a cubic surface X C lP 3 on which there are only finitely many
lines. For example, if X is given in homogeneous coordinates by the
equation
(4)
then in JA3 there are no lines on X. For by writing the equation of a line
in the parametric form T; = ait + bi{i = 1,2,3) and substituting in (4)
we come to a contradiction. The intersection of X with the plane at
infinity contains three lines. Thus, in lP 19 there exist points ~ for which
dim <p-1(~) = O. By Theorem 7 this is only possible when dim <p(r3) = 19.
Applying Theorem 1 we see that <p(r3) = lP 19.
So we have proved the following result.
§ 6. Dimension 65
Theorem 10. On every cubic sUiface there lies at least one line. In the
space lP 19 , whose points correspond to all cubic surfaces, there is an open
subset such that on sUifaces corresponding to its points there are finitely
many lines.
Cubic surfaces on which there are infinitely many lines actually
exist, for example, cubic cones. Here, too, the dimensions of fibres can
jump.
5. The Chow Coordinates of a Projective Variety. One of the most

important applications of the theorems on the dimension of intersections
lies in the fact that it enables us to specify subvarieties X C lPN of a fixed
dimension n by coordinates. We have already seen this in the case of
hypersurfaces: hypersurfaces determined by forms of degree m in JPn
correspond to points of the projective space lP'm.". Another example is
the classification of lines in lP 3 by their Plticker coordinates.
Naturally, we try to reduce somehow the case of an arbitrary variety
X to that of a hypersurface, and for this purpose we attempt to associate
with X a hypersurface. Suppose. for example, that X is a curve in lP 3 . We
consider the set Y of all lines in 1P 3 that intersect X. It is not hard to verify
that Ycorresponds to an algebraic subvariety Y on the variety fl which
describes all the lines in 1P3. Since the set y" of all lines passing through
a given point x E X corresponds. as is easy to see, to a two-dimensional
subvariety t efland dim X = 1, it is easy to deduce from Theorem 7
that dim Y = 3. Therefore Y has codimension 1 in n. If we knew that in
fl, as in lPn, any subvariety of codimension 1 is given by a single equation,
then the coefficients of this equation could be taken as coordinates of X.
The difficulty arising here can be avoided if we consider instead of lines
intersecting X pairs of planes E l ' E2 such that E 1 n E2 n X is not empty.
Since planes in lP 3 correspond to points of lP 3 (because V 3 • 1 = 3), the
pair E 1 , E2 corresponds to points of the variety 1P 3 x lP 3 of dimension
6. The set of pairs for which E I n E 2 n X '1= 13 corresponds to a subvariety
Y C lP 3 X lP 3 of dimension 5, and now we can apply Theorem 3'. This
plan will be carried out in detail for arbitrary varieties X C lPn.
e
The set of all hyperplanes C lPN corresponds one-to-one to the
points of an N-dimensional projective space, which we denote by pN,
to distinguish it from the original lPN. A hyperplane ~ and the point of
JPN corresponding to it are denoted by one and the same letter.
We consider a projective irreducible variety X C JPN, dim X = n. In
the product pN x ... X pN X X, which we abbreviate by (1PN)n+ 1 X X,
'------.r-----'
n+l
we consider the subset r consisting of those systems (~(O), ... , ~(n), x) for
which the hyperplanes ~(i)(i=O, ... , n) contain the point x E X. This r is a
closed subset, and there are two regular mappings defmed on it,
<p:r -+(1PN)n+ 1 and 1p:r -+X.
Clearly 1p(r) = X. Let us determine the dimension of 1p-l(XO), where
Xo E X: The set 1p-1{f:O) consists of the systems <e(O), ... ,e(n),xq) for which
Xo E e(I). All the e E lPN for which Xo E ~ form a hyperplane lP N- 1 C lPN.
Therefore 1p-l(xo)=(1PN-l)n+l, and dim1p-l(xo)=(N-1)x(n+1). It
follows from Theorem 7 that dim r = (N - 1) x (n + 1) + n = N(n + 1) -1,
and from Theorem 8 that r is irreducible.
There exist points y E (JPNr + 1 such that Y E cp(r) and cp - 1(y) is a
single point. This is immediately obvious from the construction
process for the sequence (1) in §6.2, if we take for the Fi linear forms that
vanish at some point x E X. Now we can apply Theorem 7 and deduce
that dimcp(r) = dimr = N(n + 1) - 1. Since cp(r) C (1PN)n+ 1 and
dim (1PN)n + 1 =N(n+ 1), we can apply Theorem 3'. It follows that cp(r)
is defined by a single form F x in n + 1 variables. The form Fx is homo-
geneous in each system of variables. Obviously we can choose it so that
it does not contain multiple factors. Then it is determined by X uniquely
to within a constant factor. F x is called an associated form, and its
coordinates the Chow coordinates of X. Let us show that the variety X, in
turn, is uniquely determined by the form F x. For this purpose it is
sufficient to verify that a point x E jpN is contained in X if and only if
any n + 1 hyperplanes ~(O), ... , e(n) containing it satisfy the relation
Fx(~(O), ... ,~(n»)=o. (1)
For if x E X, then (1) holds by the definition of F x. But if x ¢ X, then we

can find n + 1 hyperplanes ~(O), ... , e(n) containing x such that
~(O) n '" n e(n) n X = 0, which again follows immediately from the
construction of the sequence (1) in §6.2. Such ~(O), ... , e(n) do not satisfy
the relation (1).
The forms F x have a "discrete" invariant, namely the degree, and for
a given degree "continuous" invariants, the coefficients. First of all, let
us explain the meaning of the degree of F x. More accurately, it has
n + 1 degrees do, ... , d n in each system of variables. But all these degrees
are equal. In fact, since (1) is symmetrical with respect to e(O), ... , e(n)
and this condition determines the set of zeros of F x, under a permutation
of the ~(i) the form F x can only be multiplied by a constant (which is
easily seen to be equal to 1 or - 1). Therefore all the numbers di are the
same, and we denote their common value by d.
We choose n hyperplanes 11(1), ... , l1(n) C JPN such that their intersection
with X consists of finitely many points; This is always possible by virtue
of the sequence (1) in §6.2. Let
11(1) n ... n l1(n) n X = {x(1), ... , x(C)} and x(J)=(U~): ... :uW) v= 1, ... , c).
§ 6. Dimension 67
N
If we write the equation of ~(o) in the form L V;u; = 0, then we see that
;=0
F x(~(o), 11(1), ... , l1(n)) is a form of degree d in Vo, ... , VN and that this form
vanishes if and only if for at least one j = 1, ... , c
;=0
Hence it follows that

F x(~(°l, 11(1), .... l1(n)) = rf. Il (L v;uP))'j , (2)
j= 1
where rj -;. 1 are integers. So we see that c <, d, and if F x(~(°l, 11(1), ... , l1(n))
does not contain multiple factors, then c = d. Let us show that for a
suitable choice ofthe hypersurfaces 11(1), ... ,11(n) the form Fx( ~(O), 11(1), ... ,11(n))
has no multiple factors.
Lemma. If a polynomial F(x, Y), Y = (Yl, ... , Ym), does not have
multiple factors, then either there exists a Yo such that F(x, Yo) does not
have multiple factors, or F~(x, Y) = 0.
The latter case is possible only if the characteristic p of k is positive,
and then F(x, y) = G(x P , y).
We leave the simple proof of this lemma to the reader.
From the lemma and the fact that F x(~(O), ... , ~(n)) has no multiple
factors it follows that either for some choice of 11(1), ... , l1(n) the form
F x(~(O), 11(1), ... , l1(n)) does not have multiple factors, or that all the varia-
bles v\O)(i = 0, ... , N) occur in F with exponents divisible by p. But then
by virtue of the symmetry of F x in the various groups of variables all
the variables have this property, and hence F x is the p-th power of a
polynomial, contrary to our assumption.
By arguments like those used in the proof of Theorem 4 it is easy to ver-
ify that the points 11(1), ... ,I1(n)E(lPNt for which the form Fx(~(O),I1(l), ... ,I1(n))
does not have multiple factors form a non-empty open set in (lPNt
°
The points for which F x(~(O), 11(1), ... , l1(n)) i= have the same property.
Hence in (2) we have rj = 1 and c = d for points of a certain non-empty
open set.
The result we have obtained leads to the following characterization
of the degree d of the associated form of a variety X: d is the maximum
number of points of intersection in X n E, where E is a linear subspace,
dim E = N - n, and X n E is finite. The number d is called the degree of X
is denoted by deg X.
The set of all forms F(~(O), ... , ~(n)) in n + 1 groups of N + 1 variables
of degree d in each group, form a projective space IPVN.n.d, provided
that we consider forms to within a constant factor. With a variety X C IPN
of dimension n and degree d we have associated a form Fx and hence a

point c(X) in lP VN .n.d. We denote by eN n de lP VN .n.d the set of points so
obtained. The main problem is how t~ describe this set. We mention
without proof a relevant result. It asserts that CN •n •d is a quasiprojective
variety, and furthermore, it specifies the connection between X and
Fx·
To begin with we introduce an important concept. Let S be a quasi-
projective variety, r a closed subvariety of S x lPN, and cp: r -+ lPN,
lp : r -+ S its natural projections.
If for all s E S the subvarieties X s = CPlp - l(S) have one and the same di-
mension, then the family of subvarieties {Xs; S E S} is called algebraic.
We say that Sand r determine this family.
Proposition. The family of all closed subvarieties Xc lPN with
dim X = n, deg X = d, is algebraic. There exists a quasiprojective sub-
variety CN •n •d C lP"N.n.d and a closed subvariety r c CN •n •d X lPN de-
termining this family such that y E CN .n.d is the associated form for CPlp - 1 (y).
Obviously, aform FE lP"N.n'd is contained in CN •n •d if and only ifF = F x,
dim X = n, deg X = d.
A proof can be found in [18], Vol. 2, Ch. X,§8 or in [27], Ch. 1,§9.
Generally speaking, the set C N . n •d is not closed in lPVN,n.d. For example,
it is easy to see that the quadratic form F(x o, Xl> X2) is the associated
form of a conic F = 0 if and only if F is not the square of a linear form.
However, the definition can be modified somewhat so as to arrive
at closed varieties. For this purpose we introduce the concept of an
n-dimensional cycle, by which we understand a formal linear combination
D = ml Xl + ... + ml Xl of subvarieties Xi C lPN of dimension n with
integers m i > O. We set
The associated form of a so-defined cycle has all the properties of

the associated form a subvariety, but in addition, the set EN •n •d of all
associated forms of cycles D C lPN, dim D = n, deg D = d, is closed.
The concept of an associated form enables us to approach the
classification problem for subvarieties and cycles of a projective space.
Generally speaking, the algebraic variety CN,n.d is reducible. Its irreducible
components, their numbers and dimension, give a representation of the
collection of subvarieties of a given dimension and degree in lPN. We
emphasize that we consider here subvarieties in lPN not to within iso-
morphism, but that we regard them as distinct if they are distinct as sets.
In conclusion we give some examples. The case of a hypersurface
is very simple (Exercise 13). Therefore the first non-trivial examples
§ 6. Dimension 69
are curves in ]p3, that is, the case N = 3, n = 1. We list the results for
d= 1,2,3.
d= 1; C3 ,1,1 =(;3,1.1 =n is a Plticker hypersurface of the second
order. It is irreducible and of dimension 4 (see also Exercise 14).
d = 2; (;3,1,2 is reducible, (;3. \,2 = C' u C". The components C' and
C" are irreducible and dim C' = dim C" = 8. The points of C' correspond
to plane conics, the points of C" to pairs of lines (generally speaking
skew):
d = 3', E3.1.3 = C' u C" u C''' U CIV ,
dim C' = dim C" = dim C''' = dim CIV = 12.

The points of C' correspond to triples of lines, the points of C" to
reducible curves consisting of a plane conic and a line (generally speaking,
in another plane), and the points of C'" to plane cubic curves. Finally,
the points CIV correspond to twisted cubic curves. It can be shown that
all these curves are obtained from the curve V3(JP 1) (see Exercise 18) by
distinct linear transformations.
In all the cases we have been able to compute so far the irreducible
components of the variety C 3 . 1.d have not been too complicated. More
accurately, they are all rational varieties, that is, birationally isomorphic
to projective spaces. Whether this is so, in genera~ is an apparently
very difficult but very fundamental problem.
Exercises
1. Let L be an (n - 1)-dimensional linear subspace of JPn, X CLan irreducible
closed variety, and y¢L. We join y by lines to all the points x E X. We denote by Y the set
of points lying on all these lines. Show that Y is an irreducible projective variety and that
dim Y = dim X + 1.
2. Let Xc JA3 be a reducible curve whose components are the three coordinates
axes. Show that the ideal ~x cannot be generated by two elements.
3. Let X C JP2 be a reducible zero-dimensional variety whose components are
three non-collinear points. Show that the ideal ~x cannot be generated by two elements.
4. Show that any finite set of points S C 1112 can be determined by two equations
(Hint: Choose a system of coordinates x. y in JA2 so that the points of S have distinct
x-coordinates. Then define S by the equations )' =f(x), n (x - ex,) = 0, where f(x) is a
polynomiaL)
5. Show that any finite set of points S C JP2 can be given by two equations.
6. Let Xc JA3 be an algebraic curve. and x, y, z coordinates in JA3. Show that there
exists a non-zero polynomial f(x, y) that vanishes at all points of X. Show that all such
polynomials form a principal ideal (g(x, y» and that the curve g(x, y) = 0 is the closure
of the projection of X onto the (x, y)-plane parallel to the z-axis.
7. We use the notation of Exercise 6. Let h(x, y, z) = go(x, y)zn + ... + gn(x, y) be
the polynomial of smallest positive degree in z in the ideal ~x. Show that iffE ~x and if
the degree offin z is m, then!- g'O = h· U + v(x, y) and v(x, y) is divisible by g(x, y). Deduce
that the equations h = 0, 9 = 0 determine a reducible curve consisting of X and finitely
many lines parallel to the z-axis and determined by the equationsgo(x,y) =0, g(x, y) =0.
8. Using Exercises 6 and 7 show that any curve X C IA 3 can be determined by three
equations.
9. By analogy with Exercises 6-8 show that any curve X C lP 3 can be determined
by three equations.
10. Can any irreducible curve Xc IA 3 (or Xc lP 3 ) be determined by two equations?
The answer to this question is not known.*
11. Let F o(x o, ... , xn), ... , F .(xo, ... , xn) be forms of degree mo , ... , mn. Denote by r
the subset of lP n x [llP"n,m; consisting of those systems (F 0' ... , F no x) for which F o(x)
= ... = Fn(x) =0. By considering the projections <p:r->IIlP'n.m; and 1p:r->lPn show
that dim r = dim <p(n = L vn •m • - 1. Deduce that there exists a polynomial R(F 0, ... , F n)
i
in the coefficients of the forms F 0, ... , F" such that the equation R = 0 is necessary and
sufficient for the system of n + 1 equations in n + 1 unknowns F 0 = '" = F n = 0 to have a
non-zerC1'solution. What is the polynomial R if the forms F 0, ... , Fn are linear?
N
12. Show that the associated form of a point (u o : ... : UN) is of the form L ViU i •
i=O
13. Show that if X is a hypersurface and Wx=(G(uo, ... ,uN)), thenFx=G(,1o, ... ,,1N)'
where (- l)i ,1i is the minor of the matrix (vY») obtained by deleting the i-th column.
14. Let X C lP 3 be a line with the Pliicker coordinates Pi}O ,;;; i ,;;;j ,;;; 3). Show that
Fx= LPijv~O).v?).
i,j
15. Show that the degree of a hypersurface X defined by an equation G = 0 is equal
to the degree of the form G, provided that G has no multiple factors.
16. Show that subvarieties X C lP n of degree 1 are linear subspaces.
17. Let Vm: lPI -> lP m be a Veronese mapping, X = vm(lP I). Show that the degree of
X is m. Hint: Use the connection mentioned in §4.4 between hyperplane sections of vm(lPl)
and forms on lP I.
18. Show that a curve X in lP" of degree 2 either lies in a plane and is given there by
an equation of degree 2, or degenerates into two lines.
19. Let X = X I U ... U XI be a decomposition into irreducible components assumed
to be of equal dimension. Show that deg X = L deg Xi' How is the form F x connected
with the forms F x.?
20. Show that an irreducible curve X C lP n of degree d is contained in some
d-dimensionallinear subspace.
21. Show that on a Pllicker hypersurface II there lie two systems of 2-dimensional
linear subspaces. The subspace of the first system is determined by a point ~ E lP 3 and
consists of all points of II that correspond to lines L C lP 3 passing through ~. The subspace
of the second system is determined by a plane E C lP 3 and consists of all those points on
II that corresponds to line L C lP 3 lying in E. There are no other 2-dimensional linear
su bspaces on II.
22. Let F(x o, XI' Xl' X 3 ) be an arbitrary form of degree 4. Show that there exists a
polynomial <P in the coefficients of F such that the condition <P = 0 is necessary and sufficient
for the surface determined by the equation F = 0 to contain a line.
23. Let xc lP 3 be a non-degenerate quadric and Ax C II the set of points on the
hypersurface II corresponding to the lines on X. Show that Ax consistsoftwonon-intersect-
ing lines.
24. Show that the points of the space lP V'.2 corresponding to degenerate quadrics
form a hypersurface.
* The question about curves in ,13 has been answered positively under some mild assump-
tions.
Chapter II. Local Properties
§ 1. Simple and Singular Points

1. The Local Ring of a Po~t. In this chapter we study local properties
of points of algebraic varieties, that is, properties of points x e X that
are preserved when X is replaced by any affine neighbourhood of x.
Since every point has an affine neighbourhood, we may confine ourselves
in the study of local properties to points on atTme varieties.
The basic local invariant of a point x of a variety X is the local
ring (9x of this point. This ring consists of all the functions that are
regular in some neighbourhood of x. However, since distinct functions
are regular in distinct neighbourhoods, this defmition needs some care.
If X is irreducible, then (9x is a subring of the field k(X) and consists
of all the functions f e k(X) that are regular at x. Recalling the
defmition of k(X) as the field of fractions of the coordinate ring k[X],
we see that (9x consists of the ratiosflg,J,gek[X], g(x):;60.
This construction becomes clearer when we draw attention to its
general and purely algebraic character. It can be applied to any
commutative ring A and a prime ideal V in it. But here a new difficulty
arises owing to the fact that A may have divisors of zero.
Consider the set of pairs (J, g), J, g e A, g¢V, which we identify
according to the rule
(J, g) = (1', g')
if there exists an element h e A, h ¢ V, such that
h(fg'-gf,)=O. (1)
The operations in this set are defmed as follows:

(J, g) + (1', g') = (fg' + gf', gg') , (2)
(J, g) (1', g') = (Jr, gg') . (3)
It is easy to verify that in this way we obtain a ring, which is called
the local ring of the prime ideal V and is denoted by A".
72 Chapter II. Local Properties
The mapping cp:A --+ AI" cp(h) = (h, 1) is a homomorphism. The

elements cp(g), 9 ¢. p, are invertible in A p, and every element U E Ap
can be written in the form u = cp (f)/cp (g), 9 ¢. p. Occasionally the
somewhat inaccurate notation U= f/g is used. The elements of the
form cp{j)/cp(g), f E p, 9 ¢. p, form an ideal m CAp, and every element
u E AI" U ¢. m, has an inverse. Therefore m contains all other ideals of Ap.
Here we come across one of the most fundamental concepts of
commutative algebra:
A ring t!J is said to be local if it has an ideal m C t!J, m =1= t!J, containing
all other ideals.
Lemma. If A is a Noetherian ring, then every local ring Ap is also
Noetherian.
We set a = cp -1 (a) for an ideal a CAp. This is an ideal in A, which
by hypothesis has a finite basis:a = (f1' ... ,f,). If U E a, then U = cp(f)jcp(g),
9 ¢. p, 1, 9 E A. Hence it follows that f Ea, and since 1/cp(g) E A p, we
see that uEcp(u)A p =(CP(f1), ... ,Cp(!,.)). Therefore a=(cp(f1)' ... 'Cp(!,.)),
that is, a has a finite basis.
If A = k [X], where X is an affine variety, and p = mx, x E X, then
Ap is called the local ring of the point x and is denoted by t!J x. According
to the lemma, it is Noetherian.
For every pair (1, g) that determines an element of t!Jx , the function
fig is regular in a neighbourhood D(g) of x. The rule (1) indicates that
we identify in t!J x functions that coincide in some neighbourhood of x
(in our case D(h)). Thus, t!J x can also be defined as the ring whose
elements are regular functions in various neighbourhoods of x, with
the given rule of identification. Now this definition no longer depends
on the choice of any affine neighbourhood U of x.
Let us choose, in particular, a variety U whose irreducible
components all pass through x. Then a functionfthat vanishes on some
neighbourhood Vc U of x vanishes on the whole of U. Therefore the
homomorphism cp: k [U] --+ t!J x is an embedding, and we identify k [U]
with a subring of t!J x. In that case we can ignore the factor h in the
rule of identification (1). In other words, t!J x consists of functions
on U without any identifications, and all the functions cp E t!J x are of
the formflg,f, 9 E k [U], g(x) =1= 0.
2. The Tangent Space. We define the tangent space at a point x

of an affine variety X as the totality of lines passing through x and
touching X. To define what it means to say that a line L CJA.N
touches the variety Xc JA.N we assume that the system of coordinates in
JA.N is chosen so that x = (0, ... ,0) = o. Then L = {ta, t E k}, where a is a
IlXed point other than o. To investigate the intersection of X with L
we assume that X is given by a system of equations F 1 = ... = F m = 0,

with ~X=(Fl' ... , Fm).
The set X 11 L is then determined by the equations
F 1(ta) = ... = F m(ta) = O. Since we are now concerned with polynomials
in a single variable t, their common roots are the roots of their
greatest common divisor. Let
f(t) = g.c.d. (F 1 (ta) , ... , F m(ta)) ,
f(t)=Cn(t-OCi)i. (1)
The values t = OCi correspond to points of intersection of L with X.

Observe that in (1) the values t = OCi are endowed with a multiplicity
Ii, which we naturally interpret as multiplicities of the intersections
of L with X. In particular, since OeLIlX, the root t=O occurs in (1).
We arrive at the following defmition.
Defmition 1. The intersection multiplicity at a point 0 of a line L
and a variety X is the multiplicity of the root t = 0 in the polynomial
f(t) = g.c.d. (F 1 (ta) , ... , F m(ta)).
Thus, this multiplicity is the highest power of t that divides all the
polynomials Fi(ta). By definition, it is at least 1.
If the polynomials Fi(ta) vanish identically, then the intersection
multiplicity is taken to be + OC!.
Clearly,f(t) = g.c.d. (F(ta), F e ~x), so that the multiplicity does not
depend on the choice of the generators Fi of ~x.
Defmition 2. A line L touches the variety X at 0 if its intersection
multiplicity at this point is greater than 1.
Let us write down conditions for tangency of Land X. Since
o e X, the constant terms of all the polynomials Fi(T) are O. We denote
by Li their linear parts, so that Fi = Li + Gi (i = 1, ... , m), where the Gi
contain only terms of degree at least 2. Then Fi(at) = tLi(a) + Gi(ta),
and Gi(ta) is divisible by t 2 • Therefore Fi(at) is divisible by t 2 if and
only if Li(a) = o. The conditions for tangency have the form
(2)
Defmition 3. The locus of points on lines touching X at x is called

the tangent space at the point x. It is denoted by ex or, when we have
to emphasize the variety X in question, by ex x.
Thus, (2) are the equations of the tangent space. They show that
ex is a linear subspace.
Example 1. The tangent space to IN! at each of its points is the
whole oflN!. .
Example 2. Let Xc R be a hypersurface and ~x = (F). If 0 EX

and F = L + G (in the previous notation), then eo is defined by the
single equation L(T1 , ••• , T,,) = O. Therefore, if L =1= 0, then dim eo = n - 1,
and if L = 0, then eo = R and dim eo = n. An example of the second
case (with n = 2) is the curve with the equation x 2 _ y2 + x 3 = O.
3. Invariance of the Tangent Space. Defmition 3 above is given in

terms of the equations of the variety X. Therefore it is not obvious
that under an isomorphismf:X -+ Ythe tangent spaces of the points x
and f(x) are isomorphic (that is, have the same dimension). We show
that this is so, and for this purpose we reformulate the concept of the
tangent space so that it only depends on the algebra k[X].
We recall some defmitions. A polynomial F(Tl' ... , TN) at a point
x=(x 1, ... ,xN) has a Taylor expansion F(T)=F(x)+F1(T)+FiT)
+ ... + F,(T), where the Fi are homogeneous polynomials of degree i
in the variables 1) - Xj' The linear form F 1 is called the differential
polynomial of F at x and is denoted by dF or dxF. It is
dxF = i~1 (;~) (x) (7; - Xi) .
From the definition it follows that

dx(F + G) = dxF + dxG, dx(FG) = F(x)dxG + G(x) dxF . (1)
In the same notation we can write down the Eq. (2) above for the
tangent space of X at x in the form
(2)
or
L: (OF) (x) (7;-x )=O
N
-j i (j= 1, ... ,m), (3)
i=1 07;
where ~x = (F l' ... , F m). Suppose that g E k [X] is determined by some
polynomial G restricted to X. If we were to set dxg = dxG, the result
would depend on the choice of G, more accurately, it would be
determined only to within a term dxF, FE~x. Since ~x=(F1, ... ,FJ,
we have F= G1F 1 + ... + GmFm, and by (1) and the fact that Fi(X) = 0
we obtain dxF= G 1(x)d xF1 + ... + Gm(x)dxFm. Taking (2) into account,
we see that the linear forms dxF, F E ~x, vanish on ex; therefore, if
we denote by dxg the restriction of dxG to ex:
(4)
we associate with every function g Ek [X] a uniquely determined
linear form on ex'
DefmitioD. The linear function dxg defined by (4) is called the

differential of 9 at x.
Obviously,
Thus, we have a homomorphism dx:k[X]--+e~, where e~ is the

space of linear forms on ex' Since dxex = 0 for ex E k, we can replace the
investigation of this homomorphism by that of dx:mx--+e~, where
mx={jEk[X]; f(x}=O}. Clearly mx is an ideal of the ring k[X].
Theorem 1. The homomorphism dx determines an isomorphism of the
spaces mJm; and e;.
We have to prove that 1m dx = e~, Ker dx = m;. The first is
obvious: every linear form cp on ex is induced by some linear
function f on JA.n and dxf = cpo To prove the second assertion we assume
that x = 10, ... ,01, dxg = 0, 9 E m x' Suppose that 9 is induced by the poly-
nomial G E k[T1, ... , TN]. By hypothesis, the linear form dxG vanishes on
ex, and hence is a linear combination of the left-hand sides of the Eq. (2)
of this subspace:
dxG=A1dxFl + ... +ArndxFrn·
We set G1 = G-A1F 1 _ ... -ArnFrn' We see that G1 does not contain
terms of degree 0 or 1 in T1 , ... , TN' hence that G 1 E (Tl' ... , TN}2.
Further, G1Ix=Glx=g, and hence gE(t 1, ... ,tN}2, where ti=T;ix'
Since obviously mx = (t 1, ... , t N), this proves the theorem.
It is standard knowledge that if L is a linear space and M = L *
the space of all linear functions on L, then L can be identified with
the space of all linear functions on M, that is, L = M*. Applying this
device to our situation we obtain the following corollary.
Corollary 1. The tangent space at x is isomorphic to the space of
linear functions on mxlm;.
From this we can deduce a result on the behaviour of the tangent
space under regular mappings of varieties. Let f:X --+ Y be such a
mapping and f(x}=y. This determines a mapping f*:k[Y]--+k[X],
and obviously f*(m y} C m x, f*(m;} em;, so that the mapping
f *'m
• y Im2--+m
y x 1m2
x
is well-defined. Linear functions, like arbitrary functions, are map-

ped in the opposite direction, and since by Corollary 1 the spaces
ex,x and ey,y are isomorphic to the spaces of linear functions on mxlm;
e e
and my/m;, respectively, we arrive at a mapping X,x --+ y, y. This is
called the differential mapping of f and is denoted by dJ.
It is easy to verify that if g: Y--+ Z is another regular mapping and

z=g(y), then for the mapping d(gof):ex,x--+ez,z the relation
d(g f) = dg df holds. If f is the identity mapping X --+ X, then dxf
0 0
is the identity mapping of ex for every point x E X. From these

remarks we derive the following corollary.
Corollary 2. Under an isomorphism of varieties the tangent spaces
at corresponding points are mapped isomorphically. In particular, the
dimension of a tangent space is invariant under isomorphisms.
Theorem 2. The tangent space ex,x is a local invariant of x in X.
We show how to define ex in terms of the local ring (!)x of x.
We recall that the differential of a rational function FIG,
F, GEk[T1 , ••• , T"J,isdefmedas
dx(FIG) = (G(x)dxF - F(x) dxG)/G 2 (x) , G(x):;6 O.
We can regard a function f E (!) x as the restriction to X of a rational
function FIG and define the differential as dxf= dx(FIG)19 x • All the
arguments preceding Theorem 1, and also its proof, remain valid, and
we find that dx determines an isomorphism dx:mxlm;--+ei, where
now mx denotes the maximal ideal of (!)x:mx= {fE(!)x;f(x)= O}.
This proves Theorem 2.
We define the tangent space ex at a point x of any quasiprojective
variety X as (mx/m;)*, where mx is the maximal ideal of the local
ring (!) x of x. By Theorem 2 it is also the tangent space at x of any of
its affine neighbourhoods.
Thus, the tangent space is defined as an "abstract" vector space,
not specified in the form of a subspace of some larger space. However,
if X is affine and X CJAN , then the embedding i of X in JAN determines
an embedding di of 8 x,x in eX,AN. Since ex,AN can be identified with
JAN, we can regard ex x as embedded in JAN and revert to the defmition
given in § 1 . 2 . ' -
If X is projective and X ClP N, x E X and x EJAf, then ex,x is
contained in iNj. The closure of ex,x in lPN does not depend on the
choice of the open affine set iNj. Although here one and the same term
refers to two distinct objects; Bx,xClPN is also occasionally called the
tangent space to X at x.
The invariance of the tangent space permits us to answer some
questions on embeddings of varieties in affine spaces. For example,
if a point x E X is such that dim ex = N, then X is not isomorphic
to any subvariety of the affine space IAn with n < N: an isomorphism
f:X --+ YCJAn would carry ex into an isomorphic space ef(x)cN.
Starting out from this we can construct for any n> 1 an example of a
curve X CIAn that is not isomorphic to any curve YClAm with m < n.
Indeed, let X be the image of JAI under the mapping

(6)
It is sufficient to prove that ex x = N for x = (0, ... ,0). This means that
none of the polynomials F E
n
mx contains linear terms in ~, ... , T,..
Let FE~x and F = L
i=1
a/T;+G, GE(~, ... , T,,)2. Substituting (6) in F
n
we see that L aitn+i-l+G(tn,tn+l, ... ,t2n-l)=0, identically in t. But
i=1
this is impossible if at least one ai # 0, because the terms aitn+i-1 are
of degree ,,;; 2n - 1, whereas terms arising from G(t n, ... , t 2n - 1) are of
degree ;;;. 2n and cannot cancel.
From the preceding proof it follows that no neighbourhood of a
point x on a curve X is isomorphic to a quasiprojective subvariety of
JAm with m < n.
4. Singular Points. We now explain what can be said about the
dimensions of tangent spaces of points of an irreducible quasiprojective
variety X. Our result is of local character, and therefore we restrict
our attention to affine varieties.
Let X C JAN be an irreducible variety. In the direct product JAN X X
we consider the set e of those pairs (a,x), aEJAN, XEX, for which
a E ex. The Eq. (2) of § 1.3 show that e is closed ill JAN X X. We denote
by n the projection e-X:n(a, x)= x. Obviously n(e)= X, n- 1 (x)
= {(a, x); a E ex}. Thus, e is stratified into the tangent spaces to X
at various points x E X. The variety e is called the tangent fibering of X.
By § 1.3 (3) the dimension of ex is N - r, where r is the rank of the
matrix ((oF)oT;) (x»). We denote by {! the rank of the matrix (oFj/oT;)
whose elements belong to k[X]. Then all the minors of this matrix
of order greater than {! vanish, and there exist non-zero minors
Lla of order {!. Hence it follows that in the matrix (oF ;loTi) (x) all
minors of order greater than {! vanish, so that r,,;;{!, and r<{! for precisely
those points x for which all Lla(x)=O. Therefore there exists a number s
such that dim ex;;;' s and that points Y E X for which dim e y > s
form a proper closed subset of X, in other words, a subvariety of
smaller dimension.
Definition. Points x of an irreducible variety X for which
dim e x = s= min dim e y are called simple points; the remaining points
are called singular.
A variety for which a point x is simple is called non-singular at
this point. A variety is called smooth if all its points are simple. As we
have just seen, simple points form an open non-empty subset, and
singular points a closed proper subset, of X.
Let us consider the example of a hypersurface (Example 2 of § 1.2).

If ~x= (F), then the equation of the tangent space at x is of the form
n
L (oF/aT;) (x) (T; -
i=l
Xi) = 0.
We show that in this case s = min dim e y = n - 1. Clearly this is

equivalent to the fact that of/aT; do not vanish simultaneously on X.
For characteristic 0 this would mean that F is constant, and for
characteristic p> 0 that all the variables occur in F with exponents
that are multiples of p. But then (since k is algebraically closed) F = Fi,
and this contradicts the fact that ~x= (F). Thus, in our example
dim ex = dim X = n- 1 for simple points x E X. Next we show that
the same happens with an arbitrary irreducible variety and that the
general case can be reduced to that of a hypersurface.
Theorem 3. The dimension of the tangent space at a simple point is
equal to the dimension of the variety.
By virtue of the definition of a simple point the theorem asserts
that dim ex;;;;' dim X for all points x of an irreducible variety X, and
that the set of points x for which dim ex = dim X is open and non-
empty. Clearly, this is a local statement, and it is enough for us to
consider the case of an affine variety. We have seen that there exists an
integer s such that dim ex;;;;' s for all x E X and that the set of points for
which dim e x = s is open and non-empty. It remains to show that
s = dim X. Now we use Theorem 6 of Ch. I, § 3, which states that X
is birationally isomorphic to a hypersurface Y.
Let cp: X -+ Y be this birational isomorphism. According to the
proposition in Ch. I, § 4.3 there exist open and non-empty sets U C X
and Vc Y such that cp determines an isomorphism between them.
By the remarks made before the statement of the theorem, the set Wof
simple points of Y is open, and dim e y = dim Y = dim X for YEW.
The set Wn Vis also open and non-empty, hence so is cp-1(Wn V)C U.
Since the dimension of a tangent space is invariant under an iso-
morphism, dim ex = dim X for x E cp - 1(Wn U), and the theorem is
proved.
Now we tum to reducible varieties. For them even the inequality
dim ex;;;;' dim X ceases to be true. For example, if X = Xl U X 2,
dim Xl = 1, dimX 2 =2, and if x lies in Xl but not in X 2 and is a
simple point of Xl' then ex = 1 and dim X = 2. This is quite natural:
the components of X that do not pass through x influence the dimension
of X, but do not influence the space ex' Therefore it is natural to
introduce the following concept: the dimension dimxX of a variety X
at a point x is the maximum of the dimensions of the irreducible
components of X that pass through x. Obviously, dim X = max dimxX.
xeX
Definition. A point x of an affme variety X is called simple if

dim e x = dimxX.
From Theorem 3 it follows that dim ex~ dimxX for any point
x e X. For if Xi (i = 1, ... , s) are the irreducible components of X
passing through x, and e~ the tangent space to Xi at this point, then
dim e~ ~ dim Xi, e~ c ex, and therefore dim ex ~ m~ dim e ix~ max
dim Xi = dimxX. I
In exactly the same way it follows from Theorem 3 that the singular
points are contained in a subvariety of smaller dimension than X.
5. The Tangent Cone. The simplest invariant that distinguishes a

singular point from simple points is the dimension of its tangent space.
However, there is a much more delicate invariant: the tangent cone
to X'at a singular point x. We shall not need this concept later, therefore
we leave the detailed execution of the subsequent arguments to the
reader as a (very simple) exercise.
Let X be an affine irreducible variety. The tangent cone to X at
a point x e X consists of lines passing through x, which we defme as
an analogue to the limiting position of secants in differential geometry.
Suppose that xc.JP.!l, x = (0, ... ,0) and that JA.N is turned into a
vector space by choosing x as the origin .of coordinates. In
JA.N+l=JA.NxJA.l we consider the set X of pairs (a,t), aeJA.N, teJA.l ,
for which a: t e X. As always, we have two projections, namely
cP : X _ JA.l and tp : X _ JA.N. Clearly, X is closed in AN +1. It is easy to
see thai it is reducible' (if X ¥=.JP.!l) and that it consists of two com-
ponents: X=X 1 UX 2 ; X 2 ={(a,0); aeJA.N}, Xl is the closure of
cp-l(JA.l_(O») in X. We denote by CPl and tpl the restrictions of cP and
tp to it. l' The set tpl (X 1) is the closure of the set of points on all secants
of X that pass through x. The set 'Fx= tp1CP1 l(0) is called the
tangent cone to X at x.
It is easy to write down the equations of the tangent cone. The
equations of X have the form
F(at)=O, Femx'
Let F = F, + F,+ 1 + ... + Fm' where Fj is a form of degree j, Pi ¥= 0.
Then F(at) = t' F, (a) + ... + ~Fm(a). Since F(O) = 0, we always have
I ~ 1, and the equation of the component X2 is t = 0. It is easy to see
that the equations of 'Fx are of the form Pi = 0, Fe m:x . Here F, is
called the initial form of the polynomial F. Thus, 'Fx is defined by
equating to zero all initial forms of the polynomials of the ideal m:x '
Since 'Fx is determined by homogeneous equations, it is a cone with its
vertex at x. It is easy to see that 'Fx c ex, and 'Fx = ex if x is a simple
point.
so Chapter II. Local Properties
Let us consider the example of a plane algebraic curve X Cffi?

If 'llx = (F(x, y)) and Fl is the initial form of F, then the equation of
7;. is F/(x, y) = O. Since Fl is a form in two variables and k is algebraically
closed, Fl splits into a product of linear forms F1(x, y) = n (!XiX + fJiy)I,.
Therefore 7;. splits in this case into several lines !XiX + fJiY = O. These
lines are called tangents to X at X, and Ii are the multiplicities of these
tangents. If I> 1, then ex = ffi? The number I is called the multiplicity
of the singular point x. For 1= 2 it is called a double point, for 1=3 a
triple point.
For example, if F = x 2 - y2 + x 3 , X = (0, 0), then 7;. consists of two
lines: X + y = 0, X - Y = 0; if F = x 2 Y - y3 + X4, X = (0, 0), then 7;. consists
of three lines: y=O, x+y=O, x-y=O; if F=y2_x 3 , x=(O,O), then
y = 0 is a double tangent.
Like our original definition of a tangent space, this definition of a
tangent cone uses concepts that are not invariant under isomorphisms.
It can be shown, however, that the tangent cone Tx is invariant under
isomorphisms and is a local invariant of x.
Exercises
1. Show that the local ring of a point x of an irreducible variety X is the union
(in k(X)) of all rings k [U], where the U are neighbourhoods of x.
2. The mapping cp(t) = (t 2 , t 3 ) determines a birational isomorphism of the curve
y2 = x 3 and the line IA1. What rational functions of t correspond to functions of the
local ring (1) x of the point (O,O)?
3. The same for the birational isomorphism between /p,l and the curve (1) of
Ch. I. § 1.1.
°
4. Show that the local ring (1)x of the point (0,0) of a curve xy = is isomorphic to the
subring Q C (1)' EB (1)' ((1)' is the local ring of the point 0 on jp.,l) consisting of those
functions (f, g), f, g E (1)' for which f (0) = g(O).
5. Determine the local ring of the point (0, 0, 0) of the curve consisting of the three
coordinate axes in 1A3
6. Determine the local ring of the point (0,0) of the curve xy(x - r) = O.
7. Show that if x E X, Y E Yare simple points, then the point (x, Y) E X x Y is simple.
S. Show that if X=X 1 UX 2 • xEX 1 nX 2 , and if e x • x , e x • Xl , and e x . x, are the
tangent spaces, then e x • x ) ex,x, + ex. x,' Does equality always hold?
9. Show that a hypersurface of order 2 having a singular point is a cone.
10. Show that if a hypersurface of order 3 has two singular points, then the line
containing them lies on the hypersurface.
11. Show that if a plane curve of order 3 has three singular points, then it splits
in to three lines.
12. Show that the singular points of the hypersurface in ]P" given by the equation
F(xo,""x")=O are determined by the system of equations F(xo,""x,,)=O,
F x,(x o, ... , x"J = 0, (i = 0, ... , n). If the degree of the form F is not divisible by the
characteristic of the field, then the first equation is a consequence of the others.
13. Show that if a hypersurface X in ]P" contains a linear subspace L of dimension
r;;;, n12, then it has singular points. Hint: Choose a coordinate system so that L is given by
the equations x,+ 1 = 0, ... , x" = 0, and look for singular points contained in L.
14. For what values of a has the curve x~ + xi + x~ + a(xo + Xl + X 2 )3 = 0 a singular

point? What is then the nature of the singular points? Is the curve reducible?
15. Determine the singular points of the Steiner surface in 1P3:
xixi + x~X5 + X5XI - XOXI X2 X3 = O.
16. Determine the singular points of the dual Steiner surface inlP 3 :
XOXI X 2 + XOXI X3 + XOX2X3 + Xl X 2 X 3 = O.
17. Show that over a field of characteristic 0 the points of the space IPvn.m
(see Ch. I, § 5.2) corresponding to hypersurfaces having a singular point form a hyper-
surface in IPV".m. Hint: Use the results of Exercise 11 in Ch. I. ~ 6.
18. Show that if a curve of order 3 has a singular point, then it is rational. Hint: Use
the projection from this point.
19. Let F(xo, x(, x 2 )=0 be the equation of an irreducible curve XClPz. Consider
the rational mapping <p: X -> 1P2 given by the formulae Ui = JF/Jx,(x o, Xl' XZ), i = 0,1,2.
Show that a) <p(X) is a point if and only if X is a line; b) if X is not a line, then <p is
regular at X E X if and only if it is a simple point. The curve <p(X) is called dual to X.
20. Show that if X is a conic, then so is <p(X).
21. Find the dual to the curve x~ + xi + x~ = O.
22. Show that if a hypersurface xc IPn does not have singular points and is not a
hyperplane, then the set of linear subspaces ex' X E X, forms a hypersurface in the dual
space wn = IPvn.,.
23. Let 0 the group m~/mt+ I X is a finite-dimensional
vector space over k.
§ 2. Expansion in Power Series

1. Local Parameters at a Point. We investigate a simple point x of
an n-dimensional variety X.
Definition. Functions U l , ... , Un E @x are called local parameters at x
if Ui E mx and U l , ... , Un form a basis of the space mx/m;.
By virtue of the isomorphism dx : mx/m; -+ e~ we see that U I , ... , UII
form a system of local parameters if and only if the linear forms
dxu l , ... , dxu n are linearly independent on ex' Since dim e; = n, this in
turn is equivalent to the fact that the equations
(1)
have only the trivial solution in ex'
We may replace X by an affine neighbourhood X' of x on which the
functions U l , ... , Un are regular. We denote by X; the hypersurface
determined in X' by the equation U i = O. Let Ui be the polynomial that
determines on X' the function Ui' and let 2l i = 2l x;, 2£ = 2l x'. Then
2l i ) (2£, Ui ), and from the definition of the tangent space it follows
that e i eLi' where e i is the tangent space to X; at x, and Li C ex is
determined by the equation dxUi = O. From the fact that the system (1)
has only the trivial solution it follows that Li"# ex, that is, dim Li = n - 1,
and from the theorem on the dimension of an intersection and the

inequality dim 8 j ;;;. dim Xi it follows that dim 8 j ;;;. n - 1. Therefore
dim 8 j = n-1, and this means that x is a simple point on Xi. The
intersection of the varieties X; in some neighbourhood of it is just the
n
point x; if there were a component Y of the intersection X;, dim Y > 0,
passing through x, then the tangent space to Yat x would be contained
in all 8 j , and this again contradicts the fact that the system (1) only
has the trivial solution.
So we have proved the following proposition.
Theorem 1. If U 1 , ... , Un are local parameters at a point X, U1' ... , Un
are regular on X, and X j = V(Uj), then the point x is simple on each of the
n
Xi and 8 i =0, where 8 j is the tangent space to Xi at x.
Here we encounter a general property of subvarieties, which we shall
meet frequently later on.
Definition. Subvarieties Y1 , • •• , Y,. of a smooth variety X are said
to be transversal at a point x En Yi if
cOdimet01 8 x ,Y) = jt1codimxYi, 8= 8 x ,x. (2)
Using the inequality (4) in Ch. I, § 6.2 for the subspaces 8 x ,y; C 8 and
the inequality codime 8 x ,y;';;;; codimx Yi, we see that (2) leads to the
equality dim 8 x ,y; = dim Yi, which indicates that all the Yi are smooth
at x, and to the equality
codime n8
r
i=1
x ,y;=
r
L codime 8
i=1
x ,y;,
which indicates that the linear spaces 8 x ,y; are transversal: they have
the smallest possible intersection compatible with their dimensions.
From the inclusion n8
r
i=1
x ,Y/) 8 x ,y, where y=nYi, we obtain similarly
that Y is also smooth at x.
For example, two smooth curves on a surface having distinct
tangents at a point of intersection are transversal (Fig. 3).
Fig. 3
Thus, Theorem 1 asserts that the subvarieties V(Uj) are transversal.

Let X' be an atfme neighbourhood of x in which X j = x. Then x n
is determined by equations t 1 = ... = tN = 0 if X' C/AN and the tj are
coordinates, and nXj by the equations Ul = ... = un";' O. From Hilbert's
Nullstellensatz it now follows that (t 1 , ••• , tN)1 C (Ul' ... , uJ for some
I> O. Here (t 1 , ... , t N) and (Ul'"'' uJ are ideals of the ring k [X']. All
the more this is true for the ideals (t 1 , ... ,tN) and (u 1 , ••• ,uJ in (!Jx'
Observe that (t 1 , ... ,tN)=mx , so that m~C(ul,,,,,Un)' We now prove
a more precise proposition.
Theorem 2. The local parameters generate the maximal ideal mx
of the local ring (!J x'
As we have seen, mx = (t 1 , ... , tN)' What we have to show is that
all the tj E (u 1 , ••• , uJ. By induction on N - i we show that tj E (Ul, ... , Un'
t l' ... , tj _ 1)' Suppose that this is true for i = N, ... , I + 1. By hypothesis,
m x =(ul""'Un, t 1 , ••• ,tN)=(U 1 , ••• ,un> t 1 , ••• ,tl). (3)
From the defmition of local parameters it follows that
n
tl == L cxjuj(m~),
j= 1
CXj E k. (4)
By virtue of (3) every element of m~ can be written in the form

III u 1 + .•• + IlnUn + III tl + ••• + Illtl , Ilj, Ilj E m x •
Therefore (4) means that
n n I
tl= L CXjUj+ j=l
j=l
L IljUj + s=l
L Il~ts
or
n n /-1
(l-Il;)t/= L CXjUj+ L IljUj + L ll~tsE(Ul"",Un,ti,~··,t'-i)·

j=i j=i s=i
(5)
Since III E mx , we have 1 - III ¢ mx , and hence (1 - Ill) - i E (!J x' Therefore
it follows from (5) that tl E(Ui' ... , Un' t i , ... , tl - i ).
Note. The preceding argument proves the following general fact
about local rings.
Nakayama's Lemma. Let M be a module of finite type over a local
ring (!J with maximal ideal m. If the elements Ui,"" Un EM are such
that their images in M/mM generate this module, then u 1 , ••• , Un
generate M.
It is important to mention that the property of a point x of being
simple can be characterized by a purely algebraic property of its
local ring (!Jx' By defmition, x E X is a simple point if and only if
dimkmx/m; = dimxX. The left-hand side of this equality is defined for

every Noetherian local ring (g. The right-hand side can also be
expressed as a property of the local ring (gx. For by Corollary 1 to
Theorem 5 in Ch. I, § 6, the dimension of a variety X at x can be
defmed as the smallest integer r for which there exist r functions
U 1 , ••• , U, E mx such that the set determined by the equations
U 1 = 0, ... , U, = 0 consists in a certain neighbourhood of x of this
point only. By Hilbert's Nullstellensatz this property is equivalent to
the fact that (u 1 , ••• , u,)) m~ for some I> O.
For an arbitrary Noetherian local ring (g with maximal ideal m
the smallest number of elements U1, ... , U, Em for which (u 1, ... , U,)) m'
for some I> 0 is called the dimension of (g and is denoted by dim (g. By
Nakayama's lemma, m itself is generated by n elements, where
n = dim(l}/m(m/m 2). Therefore
dim (g ~ dim(l}/m(m/m2) .
If dim (g = dim(l}/m(m/m 2), then the local ring is called regular. We see
that a point x is simple if and only if its local ring (gx is regular. And
this is the algebraic meaning of simplicity of a point.
2. Expansion in Power Series. The method of associating power

series with the elements of a local ring (gx is based on the following
arguments. For every functionfE(gx we setf(x)=rJ.o,f1=f-rJ.o'
Then f1 E m x. Let U1, ... , Un be a system of local parameters at x. By
defmition, the elements U1, ... , Un generate the whole vector space
n
mxlm~. Hence there exist rJ.1"'" rJ.n E k such that f1 - L rJ.iUi Em;.
i=l
n n
We set f2 = fl - L rJ.iUi = f - rJ.O - L rJ.iUi' Since f2 Em;, we have
i= 1 i= 1
f2="J:.g jhj, gj' hjEm x • As above, there exist pji' YjiEk such that
n n
gj- L PjiUi Em;, hj - L YjiUiEm~.
i=l i= 1
We set ~('2;.PjiUi)('2;.yjiUi)= L rJ.'mU'Um· Then f2 - LrJ.'mu,um Em;,

J l l 1 sl,mSn
and hence f - rJ.o - LrJ.iUi - LrJ.'mu,um Em;. Continuing like this we can
,
obviously find forms Fi E k [T1 , ... , T,,], deg Fi = i, such that
f- L Fi(U
i=O
i , ••• ,uJEm~+l.
Definition. The ring of formal power series in the variables

(T1' ... , T,,) = T is the ring whose elements are infinite expressions of
the form
(1)
where F; E k[T] is a form of degree i, and the operations are

defined by the rules: if 'P = Go + G1 + G2 + ... , then
cP+ 'P =(Fo + GO) + (F1 + G1 )+(F2 + G2 )+ ... ,
cPo 'P=H o +H 1 +H2+ ... , H;= L GjF1•
j+I=;
The ring of formal power series is denoted by k [[TJ]. It contains

the field k (power series in which F; = 0 for i> 0). If i is the first suffIX
for which Fd 0, then F; is called the initial form of (1). The initial
form of a product is the product of the initial forms, therefore k [[T]]
has no divisors of the zero.
The preceding arguments enable us to assign to a function f E (9x
a power series cP = F 0 + F 1 + F 2 + ....
So we arrive at the following definition.
Defmition. A formal power series cP is called a Taylor series of the
function f E (9 x if for all I ~ 0
1
f- 81(U 1 , ••• ,u,,) Em~+ 1, 8 1= L F;.
;=0
(2)
Example. Let X = JA.1 and let x be the point corresponding to the

L IXmtm
00
coordinate value t = O. Then mx = (t), and the power series

m=O
associated with the rational function f(t) = P(t)/Q(t), Q(O)#O, is such
that
L
1
P(t)/Q (t) - IXmtm = O(tl+ 1),
m=O
that is,
P(t)-Q(t)cto IXmtm) =0(tI+ 1).
This is the usual way of finding the coefficients of the power series
of a rational function by the method of undetermined coefficients.
L tm, because
00
For example, 1/(1- t)=

m=O
1 I tl + 1
-- - L tl= - - =0(tI+ 1 ).
1-t m=O 1-t
The correspondence f-+cP essentially depends on the choice of the
local system of parameters u 1 , •.. , Un.
The arguments just given prove the following proposition.
Theorem 3. Every function f has at least one Taylor series.
So far we have nowhere used the fact that x is a simple point.
For U1, ••• 'U n we can choose any system of elements of (9x whose
images generate mJm;. Now we make use of the fact that x is a

simple point.
Theorem 4. If x is a simple point, then the function has a unique
Taylor series.
Clearly it is sufficient to show that any Taylor series of the
function f= 0 is zero. By (2) this is equivalent to the statement:
if F I(T1, ... , T,,) is a form of degree 1, U1, ... , Un are local parameters of
a simple point x, and
(3)
then
F I(T1, ... , T,,)= o.
Suppose that this is not so. Bya non-singular linear transformation
we can achieve that the coefficient of T; in the form FI is different
from O. This coefficient is equal to FI(O, ... ,0, 1), and if F I(O(l' ... , O(J # 0
(such 0(1' ... ' O(n exist as Pi # 0), then we need only carry out a linear
transformation that takes the vector (0(1' ... ' O(J into (0, ... , 0, 1). Thus,
we may assume that
F I(T1, ... , TJ = O(I:: + G 1 (Tl' ... , T,,-1) T;-l + ... + GI(I;., ... , T,,-1) ,
where 0( # 0 and Gi is a form of degree i.
From Theorem 2 of § 2.1 it follows easily that every element of the
ideal m~+ 1 can be written as a form of degree 1 in Ul' ... ' Un with
coefficients in m x • Therefore (3) can be written in the form
O(u~ + G1 (Ul' ... , Un-I) U~-l
+ ... + GI(Ul, ... , un-d
=IlUn+H 1 Ul,···,U n- l ) Un1-1 + ... + H(
I ( ) (4)
IU 1 ,···,Un-l,
where Il E mx , and Hi is a form of degree i. Hence it follows that
(0( - Il)U~ E{U 1, ... , Un-I). Since 0( # 0, we have 0( - wl:mx and (0( - 1l)-1 E@x,
therefore u~ E (Ul' ... , Un-I). So we see that V(U n)) V(u 1)n ... n V(u n- 1)·
Hence 8 n ) 8 1 n ... n8 n - 1 (where 8 i is the tangent space to V(Ui) at x),
and hence 8 1n ... n8n = 8 1n ... n8n - 1. Therefore dim(81n ... n8n)~ 1,
which contradicts Theorem 1 of § 2.1. This proves the theorem.
Thus, we have a uniquely determined mapping 'r:@x--+k[[T]],
which associates with every function its Taylor series. A simple check,
based on the definition (2) of'r, shows that 'r is a homomorphism. We
leave this check to the reader.
What is the kernel of-r? If 'r (f) = 0 for a function f E @x, then by (2)
this means that f E m~+ 1 for all I. In other words,! E m~. Therefore
we are concerned with functions analogous to those functions in
n
analysis whose derivatives all vanish at a certain point. We prove that
in our case such a function must be equal to zero. This is a consequence
of the following more general theorem, if we bear in mind that, as we

have proved in § 1.1, the ring @x is Noetherian.
Theorem 5. Let A be a Noetherian ring and a C A any ideal for which
the elements 1 + a, a E a, are not divisors of zero in A. Then a' = O. nI
Proof Let a E a' for any 1> 0 and a = (u 1 , •.• , uJ This means that
a can be represented as a = F ,(U1' ... , un), where F, E A [T1' ... , TJ is a
form of degree 1.
We consider the ideal generated by all the forms F,(T)(1 = 1,2, ... )
in the ring A [T]. Since A and hence A [T] is Noetherian, this ideal
has a finite basis, and we can choose a finite set of forms F " say
F l' ... , F m generating it. Then
m
F m+1(T)= L Gj(T)Fj(T) ,
j= 1
(5)
where Gj E A [T] is a form of degree m + 1 - i. We substitute in this

equation T1 = u1, T2 = U2' ... , T,. = Un' Since the degree of the form Gj is
positive, we have /lj=Gj(U1, ... ,Un)Eam+1-iCa. From (5) we find that
m
a = /la, /l = L /lj E a. Hence it follows that (1 - /l) a = 0, and since /l E a,
j =1
by the condition on a we have a = O. This completes the proof.
Corollary. A function f E (1) x is uniquely determined by anyone of its
Taylor series. In other words, the mapping r is an isomorphic embedding
of the local ring (1)x in the ring offormal power series k[[T]].
Nowhere in this section have we used the fact that X is an
irreducible variety. On the contrary, from Theorem 5 and its corollary
we can draw some conclusions on irreducibility.
Theorem 6. If x is a simple point, then one and only one component
of X passes through it.
We replace X by a neighbourhood U of x, X' = X - UZj' where
the Zj are all the components of X that do not pass through x. Then
k[X'] C@x' According to the corollary to Theorem 5, (1)x is isomorphic
to a subring of the ring of formal power series k[[TJ]. Since k[[TJJ
has no divisors of zero, this is also true of the ring k [X,], which is
isomorphic to a sub ring of it. Therefore X' is irreducible, as was
asserted by the theorem.
Corollary. The set of singular points of an algebraic variety X is
closed.
Let X = UX j be the decomposition into irreducible components.
From Theorem 6 it follows that the set of singular points of the variety
is the union of the sets Xi r. Xj (i =t j) and the set of singular points of
the Xi' Being the union of finitely many closed sets, it is closed.
3. Varieties over the Field of Real and the Field of Complex Numbers.
Assuming that k is the field of real or complex numbers, we show that
then the formal Taylor series of functions f E (()x converge for small
values of Tl , ... , T".
Let ~x = (Fl , .•• , F m), Xc JA.N and dimxX = n. If x E X is a simple
point, then the rank of the matrix
((8FJ8T)(x)) (i=1, ... ,m;j=1, ... ,N)
is N - n. Suppose that
1(8FJ8T) (x)1 #0 (i = 1, ... , N -n;j= n+ 1, ... , N). (1)
Let x be the origin of coordinates. Then t 1 , ... , tn (coordinates
restricted to X) form a system of local parameters of x. We denote by
X' the set of components passing through x of the variety determined
by the equations
(2)
By (1) the dimension of its tangent space (9' at x is n, and by the
theorem on the dimension of an intersection dimxX';;;. n. Since
dim (9';;;. dimxX', we see that dimxX' = n, and x is a simple point on X'.
Hence by Theorem 6, X' is irreducible. Clearly X') X, and since
dim X' = dim X, it follows that X' = X.
So we see that X can be determined in some neighbourhood of x
by the N - n Eq. (2), with (1) holding. By the implicit function theorem
(see, for example, [16], § 185) there exist a system of power series
cP l , ... , cP N -n in the n variables Tl , ••• , T" and 8 > 0 such that cP)Tl , .•. , T,,)
converges for all Ii with IIii < 8, and
(3)
where the coefficients of the power series cP l , .•• , cP N - n are uniquely
determined from the relations (3).
But the formal power series T(tn+l)' ... ,T(tN) (if tl, ... ,tn are chosen
as local parameters) also satisfy (3) and must therefore coincide with
cP l, ... , cP N _ n' so that it follows that T (tJ (i = n + 1, ... , N) converges
for 11j1 < 8 (j = 1, ... , n).
Any function f E (()x can be represented in the form
f=P(tl, ... ,tN)/Q(tl, ... ,tN), Q(x)#O
and T(f) = P(T(tl)' ... , T(tN))/Q(T(tl)' ... , T(tN))·
The convergence of the series T(f) follows therefore from standard
theorems on convergence of series.
Similarly it can be shown that if U l , ... , Un is any other system of
local parameters, then
1(8T(UJ/8T)(0, ... ,0)1#0 (i= 1, ... ,n;j= 1, ... ,n),
the Taylor series of t 1, ... , tn in terms of the local parameters Ul, ... , Un
are obtained by inverting the series T(Ui)=<P i(T1 , ••• ,TJ (i=1, ... ,n),
and therefore also have a positive radius of convergence. Hence it
follows that the series T(f),j E (!Jx has a positive radius of convergence
for any choice of local parameters.
The implicit function theorem asserts not only the existence of
convergent series 0
any point (t 1 , ••• ,tN )EX, Iti l<1'/ (i= 1, ... ,N), has the form
tn+i=<Pi(tl, ... ,tn) (i=1, ... ,N-n). It follows that the mapping
(t 1 , •• ·,tN )-(t 1 , ••• ,tJ carries the set (t 1 , ••• ,tN )EX, Iti l<1'/, one-to-one
and bicontinuously onto a domain of the n-dimensional space.
The space ]pN over k (when k is the field of real or of complex
numbers) is a topological space. An algebraic variety X in this space
is also a topological space. We call the relevant topology in X real
or complex according as k is the field of real or complex numbers. It
must not be confused with the topological terms such as closure,
openness, ... , which we have used earlier.
The preceding arguments show that in the real topology of an
n-dimensional variety X any simple point has a neighbourhood
homeomorphic to a domain of the real n-dimensional space. Therefore,
if all the points of X are simple, then X is an n-dimensional manifold
in the topological sense. If k is the field of complex numbers, then a
simple point x E X has in the complex topology a neighbourhood
homeomorphic to a domain in the n-dimensional complex, and so in
the 2n-dimensional real space. Therefore, if all the points of X are
simple, then X is a 2n-dimensional manifold.
As is easy to show, the space pN is compact both in the real and
complex topology. Therefore, if X is projective, then it is compact.
If k is the field of complex numbers, the converse is also true: a quasi-
projective variety X that is compact in its complex topology is a
projective variety. See Ch. VII, § 2, Exercise 1.
In conclusion we mention that everything we have said here
(excluding the last paragraph) carries over word-for-word to the case
when k is a field of p-adic numbers.
Exercises
1. Show that the set of points in which n given functions on an n-dimensional
variety X do not form a system of local parameters is closed.
2. Show that a polynomial fEk[TJ =k[ffi..1J is a local parameter at a point T=(1.
if and only if (1. is a simple root of it. .
3. Show that a formal power series e = F 0 + F 1 + ... has an inverse in k [[TJ] if and
only if F 0 "# O.
4. Consider the ring k {T} conslstmg of the expressions of the form

cC n T- n +lX_ n + 1 T- n + 1 + ... +lXo+lXl T+ ... , where T is a variable and n an arbitrary
integer. Show that k{T} is a field isomorphic to the field of fractions of the ring k[[TJ].
5. Let xc JA2 be the circle given by the equation X 2 + y2 = 1, and x the point (0, 1).
Show that X is a local parameter at x and that
r(Y)= 1 1(1--1···
2:: (_I)n,_
00 ) (1--n+l )X 2 n.
n~O n. 2 2 2
The characteristic of the ground field is O.
6. Show that if x is a singular point, then any function f E (!) x has infinitely many
distinct Taylor series.
7. Let x=N, XEX. Show that r«(!}x) does not coincide with the whole ring k[[TJ].
§ 3. Properties of Simple Points

1. Subvarieties of Codimension 1. The theory of local rings enables
us to establish an important property of a smooth variety similar to
Theorem 3 in Ch. I, § 6. It concerns the question whether a subvariety
Yc X of codimension 1 can be determined by a single equation.
Generally speaking, this is not so. (Note 2 after Corollary 5 in
Ch. I, § 6.2). We show, however, that on non-singular varieties it is true
locally. To state this result we introduce the following defmition.
Functions fl' ···.fm E (!)x are called local equations of a subvariety
Yc X in a neighbourhood of x if there exists an affine neighbourhood
X' of x such that or = (fl, ... ,fJ in k[X'], where Y' = YnX',;; E k[X'],
It is convenient to reformulate this concept in terms of the local
ring (!) x of x. For this purpose we consider the ideal ox, y C (!) x
consisting of the functions fE (!)x that vanish on Y in some neigh-
bourhood of x.
Clearly, for an affine variety X
0x,y= {f= ujv; u, v Ek[X], v(x) # 0, u E Oy},
and if all the components of Ypass through x, then Oy=ox,ynk[X].
Lemma. Functions fl, ... .fm are local equations of Y in a neigh-
bow'hood of x if and only if ox. y = (fl' ... .fm)·
It is obvious that if nY' = (fl, ... .fJ in k[X,], then also Ox y= (fl'''' .fm)
in (!)x' Let 0x,y= (fl' .. ·.fm),;; E (!)x, Oy= (gl' ... , gs), gi Ek[X].
Since gi E ox, y, we have
m
gi = L hijij
j= 1
(i = 1, ... , s), hij E (!)x' (1)
The functions;; and hij are regular in some principal affine neigh-
bourhood U of x. Let U=X - V(g), gEk[X]. The ring k[U] consists
of the elements of the form u/gl , uEk[X], 1~0. Then by (1)

(gl' ... , gJ = (ly. k[U] C (fl, .. ·,fm)·
We show that (lyk[U] = ny'. It then follows that (ly' C (fl, ... ,fJ,
and since /; E (ly', this will prove the lemma.
It remains to verify that (lyk[U] = (ly'. The inclusion (lyk[U] C (lY'
is obvious. Let VE(ly" Then v=u/gl , uEk[X], and hence U=vgl;
consequently, UEny , and since 1/gI Ek[U], we have V=U/gIE(lyk[U].
Our aim is to prove the following result.
Theorem 1. An irreducible subvariety Y C X of codimension 1 has

one local equation in a neighbourhood of any non-singular point x E X.
The proof follows precisely the lines of the proof of Theorem 3 in
Ch. I, § 6. However, there we used the unique factorization in k[T].
Here the role analogous to this ring is played by (9x, which has a similar
property.
Theorem 2. In the local ring of a simple point decomposition into

prime factors is unique.
We now prove Theorem 1, assuming Theorem 2 to be true. In
§ 3.3 we return to the proof of Theorem 2.
As we said above, the proof of Theorem 1 is the same as that of
Theorem 3 in Ch. I, § 6. Since the proposition has local character,
we may take X to be affine. Let f be any function in (9x that vanishes
on Y. We decompose it into prime factors in (9x' Owing to the
irreducibility of Yone of the prime factors must also vanish on Y. We
denote it by g and show that it is a local equation of Y. Replacing X by
a smaller affine neighbourhood, if necessary, we may assume that g is
regular on X.
Since V(g») Y and both subvarieties are of codimension 1, we have
V(g) = Yu Y'. If x E Y', then there exist functions h andh' such that
h . h' = 0 on V(g), while h =F 0 and h' =F 0 on V(g). This means that
(hh')' for some r>O is divisible by g in k[X], and a fortiori in (9x'
From the unique factorization in (9x it follows that then h or h' are
divisible by g in (9 x' Hence h or h' vanish on V(g) in some neighbourhood
of x, and after going over to a smaller neighbourhood on the whole of
V(g). This contradicts the condition. Thus, x¢: Y', and again after
replacing X by a sufficiently small affme neighbourhood of x we may
assume that V(g) = Y. If now u vanishes on Y, then US for some s> 0
is divisible by g in k [X], and hence also in (9 x' From this it follows
that u is divisible by g in (9 x' Thus, (lx, y = (g), and the theorem is
proved.
Theorem 1 has many applications. Here is the first of them.
Theorem 3. If X is a smooth variety and X _lP n a rational

qJ:
mapping of it into a projective space, then the set of points at which qJ
fails to be regular, has codimension not less than two.
We recall that the set of points of non-regularity of a rational
mapping is closed. The assertion of the theorem is of local character,
and it is sufficient to verify it for some neighbourhood of a simple point
XEX. We may write qJ in the form qJ=(fo: ... :fJ, hEk(X), and
without changing qJ we can mUltiply h by a common factor such that all
the h lie in, but have no common factor, in (r)x' Here qJ can fail to be
regular only at points where fo = f1 = ... = f" = O. But no variety Yof
codimension 1 is contained in the set dermed by these equations. For by
Theorem 1, Ux,Y = (g), and all the h would have the common factor g
in (r)x, against the hypothesis. This proves the theorem.
Corollary 1. Every rational mapping of a smooth curve into a

projective space is regular.
Corollary 2. If two smooth projective curves are birationally isomorphic,

then they are isomorphic.
Let k be the field of complex numbers. From Corollary 2 it follows
that the set of points of two curves X' and X" are homeomorphic in
their complex topology if X' and X" are birationally isomorphic. For
regular functions, and hence also regular mappings, are determined in
this case by convergent power series and are therefore necessarily
continuous.
The same is true for the sets of real points of curves defined by
equations with real coefficients if a birational isomorphism qJ : X - X'
is defined over the field of real numbers, in other words, is given by
formulae with real coefficients. From this it is sometimes easy to
conclude that two curves are not birationally isomorphic over the
field of real numbers. For exampie, the curve y2 = X3 - x has a graph
(Fig.4) consisting of two components. Therefore it is not rational
(over the field of real numbers): IPi is homeomorphic to a circle and
consists of a single component.
It can be shown on the basis of a similar idea that the curve X
with the equation y2 = X3 - x is non-rational even over the field of
complex numbers. To see this we have to compare the topological
spaces of complex points on X and on lP i in their complex topology
and to show that they are not homeomorphic. In fact, the first space is
homeomorphic to a torus, and the second to a sphere. This is a special
case of results that will be proved in Ch. VII, § 3. Figure 5 shows how the
real points of X are situated among its complex points.
"real points
_---,,-'"c..."__ \
\
\
-1 x \
Fig. 4 Fig. 5
2. Smooth Subvarieties. Theorem 1 does not generalize to subvarieties

of codimension greater than 1 (see, for example, Exercise 2 in Ch. I,
§ 6). But for subvarieties that are not singular at x, an analogous
proposition is true. We prove a somewhat more precise fact and begin
with an auxiliary proposition.
Theorem 4. Let X be an affine variety, x a simple point of it, and

u1 , ..• , Un regular functions on X forming a system of local parameters
at x. Then the subvariety Y defined by the equations u1 = ... = urn = 0
(m <; n) is non-singular at x and in some neighbourhood of x, Oy = (Ul' ... , urn),
and Urn + 1 , ... , Un form a system of local parameters at x on Y.
The proof is by induction on m. For m= 1 Theorem 1 shows that
Oy = (f) in some affine neighbourhood of x. Let U 1 = fv. Then
dxu 1 = v(x)dxf. Since U1 occurs in a system of local parameters at x,
we have dxu 1 =1= O. Therefore v (x) =1= 0, and hence Oy= (u 1) in a smaller
open set. Since dxu 1 =1= 0, we see that x is a simple point of Y.
Clearly, the tangent space ex. y to Y at x is obtained from ex, x
by imposing the condition dxu 1 = O. Therefore dxu2, ... , dxu n is a basis
of e~, y, that is, u2 , ... , Un are local parameters at x on Y.
In the general case we set X' = X Ul ' Then Y is defined on X' by the
equations U2 = ... = Un = 0, and we can apply the induction.
Now we show that any subvariety Ythat is non-singular at x can be
obtained by the process described in Theorem 4 in some neighbourhood
of a simple poiint.
Theorem 5. Let X be a variety, YC X, and x a simple point on Y

and on X. Then we can choose a system of local parameters U 1 , ... , Un
at x on X and an affine neighbourhood U of x such that Oy = (u 1 , ... , urn).
Proof To an embedding of the tangent spaces ex,y-ex,X

there corresponds an epimorphism of the associated spaces
<p: mx,x/m;,x-mx,y/m;,y, which is determined by restricting the func-
tions from X to Y. We can choose a basis U 1 , ••• , Un in mx,x/m;,x such
that Ul"'" Um E tty, and um + l' ... , Un> restricted to Y, form a basis in
mx y/m; y. We now consider an affme neighbourhood of x in which all
th~ Uj a;e regular, and in it the subvariety Y' defmed by the equations
Ul = ... = Um = O. By construction Y') Y. We show' that Y' = Y, from
which the theorem follows by virtue of Theorem 4.
According to Theorem 4, Y' is non-singular at x, hence by
§ 2 Theorem 6 Y' is irreducible in a neighbourhood of x. From
Theorem 4 it follows that dim Y' = n - m. From the construction it is
clear that dim ex y = n - m, and hence dim Y = n - m. Therefore Y = Y',
and since by The~rem 4 ttY' = (Ul' ... , u",), we also have tty = (u 1 , ... , um )
in some neighbourhood of x. This completes the proof.
In the special case X =JA.m, when k is the field of real or complex
numbers, we have already proved the analogous fact in §2.3.
3. Factorization in dte Local Ring of a Simple Point. Our proof of

Theorem 2 is based on the embedding T : (!) x - k [[TJ], where k [[T]] is
the ring of formal power series in the n variables (~, ... , TJ = T.
To begin with we mention some properties of the ring k [[T]] and
the embedding T. A formal power series must not be regarded as the
sum of its terms if the structure of k [[ T]] only is to be used: in this
ring the sum of infinitely many terms is not defmed. To make it possible,
we introduce a notion of convergence or, what is the same, a topology
in a ring of formal power series. We denote by M the ideal of k[[TJ]
consisting of the series lP (in the notation of § 2.2, (1)) for which F 0 = O.
Clearly, M = (~, ... , TJ and MI consists of all series lP for which
F j = 0 for i < I. A topology is defmed in k [[TJ] by taking as a system
of neighbourhoods of 0 the ideals MI. In other words, a sequence of
power series lPm converges to lP if the degree of the initial form of the
series lPm-lP increases beyond all bounds together with m. This can be
written as lPm-lP or lP= lim lPm. It is easy to verify that in this
topology k [[TJ] becomes a topological ring (for the defmition and
simplest properties of topological rings see [25], § 25).
L lPm' lPmE k[[TJ],
00
The series is said to converge to the sum

m=O
I
L lPm. L lPm.
00
lP if SI-lP, where SI= In this case we write lP=

m=O m=O
For example, in § 2.2 (1) SI = Fo + ... + ~ and SI-lP, because the degree
of the initial form of the series lP - SI is 1+ 1. Therefore every formal
power series is in this sense the sum of its terms.
The image r((I)x) of (l)x is everywhere dense in k[[TJ]. For if

u l , ... " Un is a system of generators of the ideal mx by means of which
we have defined the mapping r, then r(ui) = T; and r(p(Ui,···, un))
= P(TI' ... , T,.), where p is a polynomial. Since for every power series
4>=limSz, S/Ek[TJ, we have 4>=limS/(r(ul ), ... ,r(u n)), that is, 4> is the
limit of a sequence of elements in r(l)x.
The proof of Theorem 2 is based on the unique factorization in
k[[TJJ, which has first to be established. This is a fairly elementary
fact, analogous to the corresponding result for rings of polynomials.
We only indicate the main steps of the proof. A completely elementary
proof (which does not depend on the rest of the book) can be found in
[37J, Vol. 2, Ch. VII, § 1.
A power series 4>(TI , ... , T,.) is called regular relative to the variable
T,. if its initial form (of degree m, say) contains the term Cm T,:',
cm#O.
A linear transformation of the variables TI , ... , T,. evidently induces
a ring automorphism of k[[TJ]. In particular, we can carry out a
linear transformation under which the given series becomes regular
relative to T,..
Lemma 1 (Weierstrass' Preparation Theorem). If a power series
4> E k[[TJ] is regular relative to the variable T,. and if the degree of its
initial form is m, then there exists a series U E k[[TJ] whose constant term
does not vanish, such that the series 4> U is a polynomial in T,. over the
ring k[[TI' ... , T,.-IJ]:
4>U = T,.m+RI(TI, ... , T,.-I) T,:"-I + ... + Rm(TI' ... , T,.-I),
R;(TI,· .. , Tn-I) E k[[TI , ... , Tn-IJ] .
For a proof see [37J, Vol. 2, p. 139.
Lemma 2. In a ring of formal power series decomposition of elements
into prime factors is unique.
Lemma 1 makes it possible to prove this proposition by induction
on the number of variables TI , ... , T,., by reducing it to the analogous
proposition on polynomials in T,. with coefficients in k[[TI ,· .. , T,.-IJ].
The reader can find a detailed proof in [37J, Vol. 2, Ch. VII, § 1,
Theorem 6 (p. 148).
Now we move on to the proof of Theorem 2. The usual proof of the
unique factorization for integers is based on the existence of a greatest
common divisor and is valid in any ring without divisors of zero in which
any two elements a and b have a greatest common divisor d. Instead of
the greatest common divisor one can also prove the existence of a least
common multiple m, because d = (a· b)/m. Next, the fact that m is the least
common multiple of a and b means that (a)n(b) = (m). Therefore we need

only prove that in the ring (!)x the intersection of principal ideals is a
principal ideal. For this purpose we use the fact that this property is
known to us in k[[TJ] from Lemma 2. We can establish a connection
between the ideals in (!)x and in k[[TJ] by using the embedding T. From
now on we identify the elements f E (!) x with T(f). For every ideal
a C (!) x its closure a in k [[ TJ] is dermed as the set of power series tP
that are limits of sequences T(fz), fz E a.
By what we have said above, Theorem 2 is a consequence of the
following relationships between the ideals a and a.
1. anb=anb.
2. The ideal a is principal in k [[ TJ] if and only if a is principal in
(!)x'
A proof of these propositions can be found in [37J, Vol. 2, Ch. VIII.
To help the reader with the analysis of the proof we give here a brief
sketch, referring to the book for a detailed account of the arguments
(the references are to [37J, Vol. 2, Ch. VIII). We denote the ring (!)x by A,
and k[[TJ] by A.
The main thing is the notion of completion of a module E over a
ring A. This operation is similar to the transition from A to A. For
every A-module E a topology is defined, in which the submodules
Ml E are neighbourhoods of zero, and we construct a topological
A-module E in which E is embedded as an everywhere dense set and in
which the Cauchy convergence criterion holds: if an E E is a sequence
of elements such that an - am~O for n, m~ 00, then an converges to a
limit in E (this property is called completeness). The module E is con-
structed from the Cauchy sequences, that is, the sequences {an}, an E E
for which an - am ~ 0 as n, m ~ 00. Here two sequences {an} and {Pn} can
be identified if an - Pn ~ O. Every homomorphism f: E ~ F ex tends to a
homomorphismE4ft of the completions. In the special case E=A( = (!)J,
the completion E coincides with A( = k[[TJ]). If E is an ideal a C A,
then E coincides with its closure a in A. If E is a module of finite type
over A, then E is generated as an A-module by the subset E:
(1)
([37J, § 2, Theorem 5).
The following is a fundamental property of completions: if a sequence
E -4 F -4 G of finite A -modules and mappings is exact, then the same is
true for the sequence E~ ft -+ G ([37J, § 4, Theorem 11). This property
is expressed by saying that the ring A is flat over A.
To prove 1 we begin by verifying the analogous property
(2)
It follows at once from (1). Now we consider the exact sequences

0--+ anb--+ b--+b/a n b--+O,
O--+a--+a+ b--+(a+ b)/a--+O.
Since A is flat
0--+ -
an- --
over A, we obtain from them the exact sequences
b --+ b --+ b/an b --+ 0 ,
O--+o--+a+ b--+((a+b)7a)--+O,
---- -----
The isomorphism theorem b/a b ~ a + b/a shows that also
b/(an b) ~(a+b)/a,
The isomorphism theorem applied now to 0 and Ii and (2) yield the
isomorphism Ii/an b ~ Iijan Ii, from which it follows very easily that
onIi=anb.
To prove 2 we assume that a = (IX). Since a = aA by (1), we have
IX= L ai~i' (3)
and further, since ai E a,
(4)
Substituting (4) in (3) and cancelling a we fmd that L ~i 1]i = 1. Therefore
not all the 1]i are contained in M, and if 1]i ¢ M, then 1]i- 1 E A and
0= (aJ Thus, 0 = aiA, and from the fact that A is flat over A it now
follows easily that also a= aiA.
Exercises
1. Show that if t is a local parameter of a simple point of an algebraic curve, then
every function f E (lI x can be represented uniquely in the form f = tnu, where n;;. 0 and u
is an invertible element in (lI x' Hence derive Theorem 2 for curves.
2. Prove the converse of Theorem 1 in § 2: if subvarieties D " ... , Dn of codimension 1
intersect transversally at a point x and if u ... , Un are their local equations in a neighbour-
"
hood of this point, then U " ... , Un form a system of local parameters at x.
3. Is Corollary 2 to Theorem 3 true without the assumption of smoothness? Is
Theorem 3 true without this assumption?
4. Show that a point x of an algebraic curve X is simple if and only if it has a
local equation.
5. A cone Xc ,[:.,3 is given by the equation x 2 + y2 - 22 = O. Show that its generator
L given by the equations x = 0, y = z does not have a local equation in any neighbourhood
of the point (0, 0, 0).
6. A rational mapping !P: IP 2 .... IP2 is given by the formula (xo: x, : x 2 )
= (x, X2 : XOX2 : XOX,), Let x = (1 : 0: 0) and C C IP2 be a curve that is non-singular at x.
According to Theorem 3 the mapping !P restricted to C is regular at x and therefore
carries x into some point, which we denote by !pdx). Show that !Pc, (x) = !Pc, (x) if and only
if the curves C , and C 2 touch at x, that is, e x.c, = ex,c,.
7. ShoW that if ffJ = fig is a rational function, iff and g are regular at a simple point x,
and if the power series ,(f) is divisible by ,(g), then ffJ is regular at x. Hint: Use the
connections between ideals a in (r)x and their closures a in the ring of formal power series.
8. Let X eJAn and YeN' be affine varieties passing through the origins of coordinates
OeJAnand O'eJAm.A system of formal power series tP 1 (T), ... ,tPm(T) (T=(T1 , ••• , TJ) is
called a formal mapping of X into Y in a neighbourhood of 0 and 0' if tPi(O) = 0 and
F(tP!, ... ,tPm)Eaxek[[T]] for all FEay. A composition offormal mappings is defined
by substitution of the series. Two formal mappings (tP 1 , ••• ,tPml and (1pl, ... ,1pml are
called equal if tPi - 'Pi E ax (i = 1, ... , m); X and Yare called formally isomorphic in a
neighbourhood of 0 and 0' if there exist formal mappings ffJ=(tP 1 , ••• , tPml of X into Yand
1p = (1pl, ... , 1pn) of Y into X such that 1pffJ and ffJ1p are equal to the identity mappings.
Show that if the origin of coordinates is a simple point of X, then X is formally
isomorphic to an affine space.
9. Show that a formal isomorphism of N with itself (an automorphism) in a
neighbourhood of 0 is given by series tP l , ... , tPn without constant terms such that the
determinant formed from the linear terms does not vanish.
10. Show that two plane curves with the equations F = 0 and G = 0 passing through
the origin of coordinates 0 E JA2 are formally isomorphic in a neighbourhood of 0 if and
only if there exists a formal automorphism of JA2 given by series tPl and tP2 such that
F(tP 1 , tP 2) = G . U, where U is a power series with a non-vanishing constant term.
11. Show that all plane algebraic curves having the origin of coordinates 0 as a
double point with distinct tangents are formally isomorphic in a neighbourhood of 0 to
the curve with the equation xy = O. Hint: Use Exercise 10. Look in tPl and tP2 for the
highest power of the ideal (x, y).
12. Give a formal classification of double points of plane algebraic curves over a
field k of characteristic O.
13. Let X be a hypersurface in N with the equation F = F 2(T) + F 3(T) + ... + F,(T),
where F 2(n is a quadratic form of rank n. Show that X is formally isomorphic in a
neighbourhood of 0 to the cone T12 + ... + T,.2 = O.
14. Denote by k[[X]] the ring k[[TJ]/iix. Show that k[[X]] is the completion of
the local ring at x. Show also that X and Yare formally isomorphic if and" only if
k[[X]] and k[[YJ] are isomorphic.
15. Construct an embedding ,:(r)x-->k[[X]] and show that the connections
between (r) x and k [[ T]] introduced in § 3.3 also hold between (r) x and k [[ X]], even
if x is a singular point.
16. Construct infinitely many smooth projective curves that are pairwise non-
isomorphic to each other over the field of real numbers.
§ 4. The Structure of Birational Isomorphisms

1. The tr-Process in a Projective Space. In the preceding section we have
shown (Corollary 2 to Theorem 3) that a birational isomorphism
between smooth projective curves is an isomorphism. For varieties of
higher dimension this is no longer true: for example, stereographic projec-
tion, which establishes a birational isomorphism between anon-degenerate
quadric and a projective plane, is not a regular mapping (Exercise 9 of
Ch. I, § 4, and the remark after Corollary 5 to Theorem 4 in Ch. I, § 6).
In this section we define and investigate the simplest and most typical
birational, but not regular, isomorphism: the a-process.
We consider the projective spaces pn with the homogeneous coordi-

nates xo, ... , Xn and pn-l with the homogeneous coordinates Yl' ... , Yn'
In the space pn X pn-1 we denote a point x x y, X= (xo: ... : x n),
Y= (Y1 : ... : yJ also by (xo: '" : xn; Yl : ... : yJ. We consider the closed
subvariety II C pn X pn-1 defmed by the equations
(1)
Defmition 1. The mapping u: II ~ pn defmed by the projection

pn X pn-l ~ pn is called the u-process.
Let ~ denote the point (1 : 0 : ... : 0) E pn. If (Xo : ... : xJ #:~, then it
fo llows from (1) that (y 1 : ..• : Yn) = (X 1 : .•. : xJ, so that the mapping
(2)
is inverse to u. But if (Xo : ... : xJ = ~, then arbitrary values of Yi satisfy

the equations. Thus, u- 1 W= ~ X pn -1, and u determines an isomor-
phism between pn_~ and II_(~xpn-l). The point ~ is called the
centre of the u-process.
Now we describe the structure of II in a neighbourhood of points
of the form (~; Y1 : ... : yJ. We have Yi #: 0 for some i, consequently the
chosen point lies in the open set Ui determined by the conditions Xo #: 0,
Yi#:O. In this set we may even assume that Xo= 1, Yi= 1. The Eqs. (1)
then take the form x j= Yjx i, 1 ..;.j#: i..;. n. Hence it follows that Ui is iso-
morphic to an affme space with the coordinates Yl' ... , Xi' ... , Yn.
In particular, we see that II is not singular and hence, by
Theorem 6 of § 2, is irreducible in a neighbourhood of each of its points.
We shall soon see that II is irreducible.
To give a clearer picture of the action of a u-process, we consider
it on some line L passing through~. Let x j =(1.jXi (j=1, ... ,n,j#:i,
i #: 0) be the equations of this line. On L the mapping (2) takes the form
u- 1(xo: '" : xJ = (xo : ... : Xn ; (1.1:"': tI : ... : (1.n). We see that u- 1 is
regular on L and carries it into a curve u- 1 (L), which intersects
~ Xpn-l in the point (~; (1.1 : ... : t : ... : (1.n). We can interpret this result
I
as follows. The mapping u- 1 is not regular at ~, but by regarding it on the

line L we obtain a regular mapping u- 1 : L~ II. By using it we can
define u- 1 also at ~ (over the field of real or of complex numbers this
would mean that we define u- 1 (x) for XEL and let x tend to ~ in the
direction of L). However, the result depends on the choice of L (the
limit process depends on the directions in which we carry it out). By
choosing various lines L we obtain all possible points on ~ Xpn-1.
Thus, although u- 1 is not regular at ~, by resolving the resulting in-
determinacy we obtain not arbitrary points of II but only points of
ex pn-1. Having this picture in mind we can say that cr- 1 blows up e
to ex pn-1. _
Observe that at the same time we have proved that II is irreducible.
For
II =.(e x pn-1)u(II -(e x pn-1»).
Since II - (e
x pn-1) is isomorphic to pn - e,
it is irreducible, and
hence so is II - (e x pn 1). What we have to verify is that
expn-1CII-(expn 1).
But necessarily
cr-1(L)CII-(expn 1),
hence
cr- 1 (L)n(e x pn-1) C II _ (e x pn 1).
We have already seen that for a suitable choice of L we can obtain any
e
point of x ]pn-1 on the left-h.and side.
For n = 2 we can give a clear illustration of the mapping cr: II -+ p2
and its action on lines L. The curves cr - 1 (L) intersect the line x P 1 e
e
at points that do not change as L rotates about in p2. Thus, II looks
like one loop of a helix (Fig. 6).
Fig. 6
2. The Local a-Process. Now we construct for any quasiprojective

variety X and a simple point x of it a variety Y and a mapping
cr: Y -+ X, analogous to that constructed above for X = ]pn.
We begin with an auxiliary construction.
Let X be a quasiprojective irreducible variety, ~ a simple point of it,

and Ul' ... , u" functions regular on the whole of X and such that:
a) the equations Ul = ... = Un = 0 have on X the unique solution ~;
b) the functions U1, ... , Un forms a system of local coordinates at ~.
We consider the product X x lP n- 1 and in it the subvariety Y con-
sisting of those points (x; t 1 : ... : t n), x E X, (t 1 : ... : t n) E lP n - 1 such that
Ui(X) t j = Uix) t i, 1 .;;; i, j .;;; n. The regular mapping (J : Y -+ X which is the
restriction to Y of the projection X x lP n - 1 -+ X is called the local
(J-process centred at ~.
Note that this construction, generally speaking, is not applicable
when X is projective: we require the existence on X of non-constant
everywhere regular functions Ul' ... 'Un' Therefore the new concept does
not comprise the earlier introduced concept of a (J-process for the case
X = lPn. The connection between them is as follows.
We denote by X the affine subset defined in lP n by the condition
Xo '# 0, and we set Y = (J - 1 (X). Then the mapping (J : Y -+ X induced on
Y by the (J-process II -+ lP n is a local (J-process.
The following properties, which we have proved in § 4.1 for the
(J-process, can be proved word-for-word in the same way for a local
(J-process: the mapping (J: Y -+ X is regular and determines an iso-
morphism
At a point Y E (J -1 (~) we have ti '# 0 for some i, and we can set

Sj=t/t i, j'#i. Then the equations of Y take the form Uj=UiSj
U= 1, ... ,n, j '# i). Hence we see that the ideal of Y has the form
my = (Ul - U1 (Y), ... , Un - Un(y), SI - SI (Y), ... , Sn - Sn(y))
= (SI- SI(Y)' ... , Ui - Ui(Y), ... , Sn - Sn(Y)) .
Therefore dim e y, y .;;; n, and since dim (J - I(X -~) = n, the variety Y
is smooth at every point Y E (J-l(X - ~). Since
Y = (J-l(X _ ~)u(~ x lPn-I) ,
Y is either irreducible and therefore coincides with the closure (J I(X -~)
of (J- I(X - ~), or it has one component isomorphic to lPn-I. In the second
case the two components intersect: otherwise (J-l (X - ~) would be closed,
but then, by Theorem 3 of Ch. I, § 5, its image X - ~ would also be
closed. The point of intersection of the two components would be simple,
and this contradicts Theorem 6 of § 2. Thus, Y is irreducible and smooth,
and SI-SI(Y), ... ,Ui-Ui(Y)' ""sn-sn(Y) are local parameters at a point
Y E (J-l (~) at which t;# O.
Now we establish a property that could be called the independence
of a local (J-process of the choice of the functions Ul' ... , Un'
Lemma. If V 1, ... , vn is another system of functions on X satisfying

the conditions a) and b), if Y' is the resulting variety, and 0"' : Y' ~ X the
corresponding local O"-process, then Y' and Yare isomorphic.
There even exists an isomorphism lp : Y ~ Y' such that the diagram
commutes.
Proof Let Y' C X x lP n -1 and let t~, ... , t~ be homogeneous co-
ordinates in lP n - 1 • In the open sets Y - 0"-1(~) and Y' - 0"'-1@ we set
lp(x; t1 : ... : tJ=(x; v1(x): ... : vn(x»),
(1)
lp(x; t~ : ... : t~)=Jx; U1(X): ... : uix»).
From property a) of the functions Ui it follows that lp and 11' are
regular and that lp(Y - 0"-1(~»C Y', lp(Y' - 0"'-1(~)C Y.
We now consider an open set in which ti ¥= 0, and we set sj = titi'
Since VI(~=O and U 1 " " ' U n is a basis of the ideal m~, we have
L hljuj,
n
VI = hljE(!)~. (2)
j=1
n
VI = Ui
j=
L1 0"* (hlj) Sj= UigZ,
n
(3)
gz = L O"*(hlj) Sj.
j= 1
We set lp(X; t1 : ... : tn) = (x; g1 : ... : gn). Evidently our mapping coin-
cides with (1) in their common domain of definition, because there
gl = V,/Ui' Let us verify that cp is regular. For this purpose we have to
show that g1' ... ,gn do not vanish simultaneously at any point
l1EO"-1(~). Suppose that all the gl(rJ}=O. Since not all the Sj(l1)=O
(because Sj = 1), it follows from (3) that Ihli~)1 = O. But
VI == I;hli~)' uj(mod m~),
and from this it would follow that the VI are linearly dependent in m~/m~,
whereas they form a system of local parameters at ~. In this way we define
a unique mapping cp: Y ~ Y', and similarly 11': Y' --+ Y. The fact that they
are inverses of one another need only be verified on an open set, where
the formulae (1) hold. But there it is obvious.
3. Behaviour of Subvarieties under a a-Process. Let X be an irre-

ducible quasiprojective subvariety of PN, and a:ll -+pN the a-process
dermed in §4.1. We investigate the inverse image CT- 1 (X) of X, which is,
of course, a quasiprojective subvariety of ll.
Theorem 1. If XCpN, X is non-singular at ~ and X#pN, then rel-
ative to the CT-process centered at ~ the inverse image a- 1 (X) is reducible
and consists of two components:
(1)
On the component Y the mapping CT: Y -+X determines a regular mapping.

It is an isomorphism between some neighbourhood U of a point x E X and
CT- 1 (U) provided that x #~, and it is the local CT-process CT- 1(U)-+U when
x=~.
Proof We denote by Y the closure CT 1(X - ~ of the set a- 1(X -~.

Since 0'-1 is an isomorphism in pN -~, we see that CT- 1 (X -~ is
isomorphic to X - ~ and hence irreducible. Consequently so is Y. From
the defiqition it is clear that (1) holds: if x E X -~, then
CT- 1 (X)EY, CT-1@=~XpN-1 .
.
The fact that CT : Y -+ X is an isomorphism in a neighbourhood of an
arbitrary point x E X, except x =~, has already been mentioned. It
remains to investigate this mapping in a neighbourhood of ~.
Here we can use the fact that in an affme space containing ~, a
CT-process can be described as a local CT-process and that a local
CT-process does not depend on the choice of local coordinates. For by
Theorem 5 of § 3 we can choose a system of local coordinates
u1 , ••• , UN at a point ~ E pN such that in some neighbourhood of this
point the variety X is given by the equations
(2)
and that the functions U1' ••• , Un determine a local system of coordinates
on X at ~. We can choose a neighbourhood U C pN of ~ so that
u1 , ... , UN satisfy the conditions a) and b) of the lemma in § 4.2, and
therefore the proof of the theorem reduces to the special case when X is
given by (2).
From the conditions a), b), and uitj=uA we deduce that
tn+ 1(x) = ... = tN(x) = 0 for x # ~. Therefore Y is contained in the subspace
Y' dermed in X x pN -1 by the equations
tn + 1 =···=tN =O, (3)
Ui tj = Uj ti , 1 "" i, j "" n . (4)
If we denote by lP n- 1 the subspace of the projective space lPN -1 defined

by (3~ then we see that Y' C X x lP n - 1 is determined by (4). Thus, Y'
coincides with the variety obtained by the local u-process. We have now
shown that Y' = u 1 (X - ~). Therefore Y = Y', and this proves the
theorem.
We can now give the most general defmition of a u-process. If X is
a quasiprojective variety, X C lPn, ~ a simple point of it, and Y the
variety introduced in the statement of Theorem 1, then u: Y ~X is called
the u-process centered at ~. From what we have proved about the local
u-process it follows that Yis irreducible if X is, that all points of U-1(~) are
simple on Y, and that U-1(~) ~ ~ x lP n- 1 •
Note that the u-process is an isomorphism if X is a curve. Thus, the
presence of a non-trivial u-process is a characteristic feature of many-
dimensional algebraic geometry.
4. Exceptional Subvarieties. The example of the u-process points to a

fundamental difference between algebraic curves and varieties of dimen-
sion n > 1. Whereas a birational isomorphism for non-singular projective
curves is an isomorphism, the u-process gives an example to show that
this need not be the case for higher dimensions.
We mention one peculiarity of the u-process: it is a regular
mapping and fails to be an isomorphism only because the rational
mapping u- 1 is non-regular (at ~).
We now investigate the mapping f : X ~ Y, where f is a regular
mapping and a birational isomorphism, that is, f -1 = g is a rational,
but non-regular mapping Y ~ X. In the example of the u-process we
have seen that a subvariety of codimension 1 in Y is contracted to a
point ~. Let us show that the analogous property always holds in this
situation.
Theorem 2. Suppose that f : X ~ Y is a regular mapping and a birational
isomorphism, that y=f(x) is a simple point on Yfor XEX, and that the
mapping g = f -1 is non-regular at y. Then there exists a subvariety
Z C X with x E Z, such that codim Z = 1, codimf(Z);;;;. 2.
Proof We can replace X, if necessary, by an affme neighbourhood
of x and may therefore assume that X is affme. Suppose that X CAN
and that g=f- 1 is given by the formulae tj=gj(i= 1, ... ,N), gjEk(Y),
where t 1 , ••. , tN are coordinates in AN.
Evidently gj = g*(t j), and since "g is non-regular at y, at least one of
the functions gj is non-regular at y. Suppose that this is g1' so that
g1 ¢ {9y. We can represent g1 in the form g1 = u/v, u, v E (9y, v(y) = 0,
and since prime factorization in {9 y is unique (by assumption y is a
simple point), we can choose u and v relatively prime. Since g=f- 1,
we have tl = f*(gl) = f*(u/v) = f*(u)/ f*(v), therefore
f*(v)t 1 = f*(u). (1)
Clearly f*(v) (x) = 0, so that x E V(f*(v»). We set Z = V(f*(v»). By the

theorem on the dimension of an intersection codim Z = 1, because
XEZ, and therefore Z is not empty. From (1) it follows that f*(u) = 0
on Z, because t 1 is a regular function. Therefore u = 0 and v = 0 on
f(z), and hencef(Z)c V(u)n V(v).
It remains to verify that codim (V(u)n V(v»);;;;. 2. But if V(u)n V(v)
contained a component Y' with y E Y', codim Y' = 1, then according
to Theorem 1 of § 3, Y' would have a local equation h. This would
mean that u E (h), v E (h), which contradicts the fact that u and v have
no common factor in (!}y.
Defmition. Let f: X - Y be a regular mapping and a birational
isomorphism. A subvariety Z C X is called exceptional if codim Z = 1,
codimf(Z);;;;. 2.
Corollary 1. If a regular mapping of smooth varieties f: X - Y is a
birational isomorphism, but not an isomorphism, then it has an exceptional
subvariety.
Corollary 2. If f: X - Y is a regular mapping and a birational
isomorphism, where X and Yare curves and Y is smooth, then f(X) is
open in Y and f determines an isomorphism between X and f (X).
The fact that f(X) is open in Y follows from the existence of
isomorphic open subsets U and V in X and Y. Since f(U)= V is
obtained from Y by removing fmitely many points, a fortiori f(X) is
obtained in this way, hence is open in Y. If the mapping f: X -f(X)
were not an isomorphism, then we would come to a contradiction to
Theorem 2, because in our case only the empty set has codimension ;;;;. 2.
5. Isomorphism and Birational Isomorphism. We consider the class

of all birationally isomorphic algebraic quasiprojective varieties. All
the representatives of this class are called its models.
In the next section we show that every class of birationally
isomorphic curves contains a projective smooth model X o. Corollary 2
to Theorem 3 of § 3 asserts that there is only one such model (to within
isomorphism). Therefore, if we associate with every class the unique
non-singular projective model contained in it, we reduce the classification
problem for algebraic curves to within a birational isomorphism to
the same problem for non-singular projective curves to within an
isomorphism.
Fields of functions on algebraic curves are fmitely generated

extensions of transcendence degree 1 of a field k. We can therefore set
up a one-to-one correspondence between such fields K and non-singular
projective curves. Under this correspondence K = k(X). We also call
X a model of K.
One could try to fmd a model X directly, starting out from
algebraic properties of K. We make this problem more precise by
asking how the local rings of all points of a curve X can be
characterized within K. It is easy to verify that every local ring ((Jx of a
point x e X has the following properties:
1) ((J is a subring of K, k~((J~K;
2) ((J is a local ring and its maximal ideal m is principal: m = (u);
3) the field of fractions of ((J is K.
It can be shown (Exercises 7, 8, 9) that any subring ((J of K having
the properties 1), 2), and 3), is the local ring ((Jx of a suitable point
x e X. Thus, X is a universal model: it contains all local rings of K
satisfying the natural conditions 1), 2), and 3).
How can one solve these problems for varieties of dimension
n> 1? When a projective non-singular model exists, things are com-
paratively well-behaved: the existence was proved for n = 2 and 3
(Walker and Zariski for fields of characteristic 0, and Abhyankar for
any fmite characteristic greater than 5), and for an arbitrary n and
~haracteristic 0 (Hironaka). The existence is highly probable for an
arbitrary field and arbitrary n. On the other hand, the uniqueness of a
non-singular projective model is an exceptional feature of the case
n = 1. This is clear from the example of the projective plane IP2 and a
quadric, which are birationally isomorphic, but not isomorphic.
One could ask whether perhaps in every class of birationally
isomorphic varieties there exists a model that is universal in the sense
that the local rings of its points, as in the case n = 1, exhaust all the
local subrings of the field K = k(X) satisfying the conditions 1), 2),
and 3) [except that in 2) m=(u1,""U,.) instead ofm=(u)]. However,
for the same reasons such models cannot exist. For if u : X'- X is the
u-process centered at eeX, then the local rings of points yeu- 1 m
do not coincide with any of the local rings ((Jx, x e X. The reader can
easily prove this as an exercise. True, by combining all the non-
singular models of one class we can obtain a certain object having this
property of universality; however, it is not a fmite-dimensional
algebraic variety. Some information on this "infmite model" can be
found in [37], Vol. 2, Ch. VI, § 17.
Owing to the absence of a distinguished model the problem arises
of studying the connections between various non-singular projective
models of one class of birationally isomorphic varieties. Without
proof we quote the relevant fundamental results. In what follows, all

varieties are assumed to be irreducible, smooth, and projective.
We begin with two terms. A model X' dominates X if there exists
a birational regular mappingf: X' -+ X.
A variety is called a relatively minimal model if it does not dominate
any variety not isomorphic to it. For example, a smooth projective curve
is always a relatively minimal model. By Theorem 2 a variety is a
relatively minimal model if it has no exceptional subvarieties.
It can be shown that every variety dominates at least one relatively
minimal model. Thus, every class of birationally isomorphic varieties
contains at least one relatively minimal model.
Now the important question of its uniqueness arises. If in every
class there were such a unique model, then again it would reduce the
birational classification to the classification to within isomorphism.
However, for n> 1 this is not the case. An example is the projective
plane ]P2 and a quadric Q, which we know to be birationally isomorphic,
so that they are tinodels of one and the same class of birationally
isomorphic surfaces. We show that ]P2 and Q are both relatively
minimal models, in other words, do not have exceptional curves.
Since ]P2 and Q are non-isomorphic (Remark 1 to Ch. I, § 6.2), this
gives the required example.
In our case an irreducible exceptional curve C C X must contract
to a point y E Y :f(C) = y under a regular birational mapping f: X -+ Y.
Here X and Yare projective surfaces. Such curves have a number of
very special properties (which explains the term "exceptional"). We
mention only one of them.
According to Theorem 3 of § 3 the mapping f - 1 fails to be regular
at only fmitely many points Yi E Y. Let U be a sufficiently small atrme
neighbourhood of y such that f - 1 is regular at all points of U other
than y. We set V=f-1(U), C=f-1(y). Obviously, Vis an open subset
of X and V) C. We show that in V there is no irreducible curve C'
closed in X and not contained in C. For C' is a projective curve and its
image f(C) is also projective. But f(C) C U, which is atrme. By
Corollary 2 to Theorem 3 of Ch. I, § 5, this is only possible when
f(C)= y' is a point. If y' # y, then C' must also be a point, becausef-l
is an isomorphism in U, apart from y. But if y' = y, then C' cf-1(y) = C.
Thus, C lies isolated in X: in some neighbourhood Vof it there are
no irreducible projective curves not contained in C. In other words,
C cannot "move just a little". From this one can derive that many
surfaces do not contain exceptional curves.
For example, let X = ]p2, V=]p2 - D) C, where C is an exceptional
curve. Then dim D = 0, because otherwise C and D would intersect by
the theorem on the dimension of an intersection. But if dim D = 0,
that is, D is a rmite point set, then there exist arbitrarily many curves C
that do not intersect D, for example straight lines.
Let X = Q. Here we use the existence of a group of projective
transformations G carrying Q into itself. We recall that transformations
of G are given by matrices A of order 4 satisfying the relation
A* FA = F, where F is the matrix of the equation of Q. Hence it follows
that G is an algebraic subvariety in the space of all matrices of order 4.
In what follows we may therefore take G to be an algebraic affine
variety.
If C is a curve and CeQ - D, then we construct a transformation
cP E G such that cp(C) ct. C, cp(C) c Q - D, but this contradicts the above
property of exceptional curves. It is enough to show that the set
of those cp E G for which cp(C)IID =F 0 is closed. Then we have at our
disposal an entire neighbourhood of the identity transformation e E G
consisting of elements with the required property. To describe the set
S of those cP E G for which cP (C) II D =F 0 we consider in the direct
product G x Q the set r of those pairs (cp, x) for which x E C, cp(x) E D.
Clearly r is closed. If f: G x Q --+ G is the natural projection, then
S = f(r), and f(r) is closed according to Theorem 3 of Ch. I, § 5. This
completes the proof that two distinct minimal models exist.
It is all the more surprising that nevertheless uniqueness of the
minimal model holds for algebraic surfaces, provided that some
special types are excluded. Namely, Enriques has shown that in a class
of surfaces the minimal model is unique if the class does not contain
a surface of the form C X]pI, where C is an algebraic curve. (Surfaces
birationally isomorphic to C X]p1 are called ruled.)
For this result see [3], Ch. II, § 4,42.
About minimal models for varieties of dimension n;;;. 3 nothing
is known.
Exercises
1. Suppose that dim X = 2, that ~ is a simple point of X, that C1 and C2 C X are two
curves passing through ~ and not singular there; a: Y.... X the a-process centered at ~,
C;=a l(C; -~), Z=a-l(~). Show that C1I1Z=C 2I1Z if and only if C 1 and C 2
touch at~.
2. Suppose that dim X = 2, that ~ is a simple point of X, C C X a curve with ~ E C,
and! the local equation of C in a neighbourhood of ~. Let!= fI (oc;u+ !3;V)I; (m~+l),
i=l
l:.1; = I, where U and v are local parameters at ~ and the forms OC;U + !3;v are not
proportional to each other.
As in Exercise 1, a: Y.... X; C' = a-i(C -~. Show that C'IIZ consists of r points.
3. The notation is that of Exercise 2, but firstly, !=(OC1U+ !31V)(OC2U+ !32V) (mn,
and secondly, the linear forms OC 1U+!31v and OC2U+!32V are not propertional. Show that
the two points of C'IIZ are simple on C'.
4. Consider the rational mapping cp : lP 2 ...... lP 4 given by the formula

qJ(xo : XI : x 2) = (XOXI : XOX2 : xf : XI X2 : x~) .
Show that qJ is a birational isomorphism and that the inverse mapping qJ(1P2)-+lP 2 is a
q-process.
5. As in Exercise 4, investigate the mapping lP 2 -+lP 6 given by all monomials of
degree 3 other than x~, x~ and x~.
6. Construct an example of a birational isomorphism X -+ Y under which an
exceptional subvariety of codimension 1 is carried into a subvariety of codimension 2
(dim X = n, where n is arbitrary).
7. Let (!J be a local ring of the field k(X) satisfying the conditions 1) - 3) of § 4.5
(X is a projective algebraic curve). Show that for every U E k(X) either U E (!J or u- I E (!J.
Let X C lPn, and let X o , ... , Xn be the homogeneous coordinates in lPn. Show that there
exists an i such that XiXjE(!J(j=O, .•• ,n).
8. The notation is the same as in Exercise 7. Let X' be an affme curve, X' = X n~.
Show that k[X1c(!J, that k[X]nm is the ideal of some point XEX', and that (!JxC(!J.
9. Show that if two rings (!J I and (!J 2 satisfy the conditions 1) - 3) of § 4.5 and
(!JI C (!J2' then (!JI = (!J2. Hence and from Exercises 7 and 8 derive that (in the notation of
Exercise 8) (!J = (!Jx.
to. Let V be the quadric cone given by the equation xy = Z2 in JA.3, and let X' -+ JA.3
-be the q-process centered at the origin of coordinates, and V'the closure of the
subvariety q-I(V - 0) in X'. Show that V' is a smooth variety and that the inverse image of
the origin of coordinates under the mapping q: V' -+ Vis a smooth rational curve.
§ 5. Normal Varieties
1. Normality. We begin by recalling one algebraic concept. A ring A
without divisors of zero is called integrally closed if every element of
the field of fractions K of A that is integral over A belongs to A.
Defmition. An irreducible affine variety X is called normal if the
ring k [X] is integrally closed. An irreducible quasiprojective variety X
is called normal if each of its points has an affine normal neighbourhood.
We shall prove presently that smooth varieties are normal
(Theorem 1). Here is an example of a non-normal variety. On the curve
X with the equation
y2= x 2 +X3
the function t = y/x E k(X) is integral over k [X], because t 2 = 1 + x,
however, t¢ k[X] (Exercise 9 of Ch. I, § 3). This example shows that the
point set of normality has some relation to the singular points of the
variety.
Our next example is a variety that has a singular point but is normal.
This is the cone X with the equation x 2 + y2 = Z2 in fA 3 (we assume that
the characteristic of the ground field is not 2). -
Let us prove that the ring k[X] is integrally closed in k(X). In
doing this we use the simplest properties of integral elements (see [37],
Vol. 1, Ch. V, § 1). The field k(X) consists of the elements of the form
u + VZ, where u, v E k(x, y) and x and yare independent variables.
Similarly k [X] consists of those elements of k(X) for which u,
vEk[x,y], so that k[X] is a fmite module over k[x,y] and hence all
the elements of k [X] are integral over k [x, y]. If oc = u + VZ E k (X) is
integral over k[X], then it must also be integral over k[x,y]. Its
minimal polynomial has the form T2 - 2uT+ (u 2 - (x 2 + y2)V 2), hence
2u E k [x, y] and u E k [x, y]. Similarly, u2 - (x 2 + y2) v2 E k [x, y], and
hence (x 2 + y2) v2 E k [x, y].
Since x 2 + y2 = (x + iy) (x - iy) is the product of two coprime
elements, we see that vEk[x,y], and this means that ocEk[X].
We now establish some simple properties of normal varieties.
Lemma. An irreducible variety X is normal if and only if the local rings
(!)x of all its points are integrally closed.
Since the defmition of normality is of local character, we may
confine our attention to the case when X is affine. Let X be normal,
x E X. We prove that (!)x is integrally closed. Let oc E k(X) be integral
over (!) x' that is,
(1)
Here ai E (!)x, and therefore ai = bi/ci, bi, Ci E k [X], ci(x);6 0. Setting

do=ct ... cn and multiplying (1) by do we fmd that
(2)
where di E k [X], do(x);6 0. Multiplying (2) by d"o-t and setting dooc = [3
we find that [3 is integral over k [X]. By hypothesis, k [X] is integrally
closed, hence doOC=[3Ek[X]. Then OC=[3/d o E(!)x, since do(x);60.
Suppose now that all the (!)x are integrally closed. We show that
then k[X] is integrally closed. If ocEk(X) and oc is integral over k[X],
then ocn+ at ocn - t + ... + an = 0, ai E k [X]. But then all the more ai E (!)x
for every x E X, and since (!)x is integrally closed by hypothesis, we see
that oc E (!) X' Therefore oc En (!) X' According to Theorem 4 of Ch. I, § 3,
n(!)x=k[X] and hence ocEk[X].

xeX
xeX
Theorem 1. Smooth varieties are normal.

By virtue of the lemma it is sufficient to show that if x is a simple
point, then the ring (!)x is integrally closed. We know that factorization
in (!) x is unique (Theorem 2 of § 3). Every element oc E k (X) can be
represented in the form oc = u/v, where u, v E (!) x have no common
divisors. If oc is integral over (!)x, then ocn+ aiocn-t + ... + an = 0, ai E (!)x'
Hence un + at un- t V+ ... + anv n= 0. So we see that vlu n. Since u and v
are coprime and factorization is unique, it follows that oc E (!)x'
Theorem 1 shows that normality is a weaker form of smoothness.

This becomes apparent also in properties of normal varieties. In parti-
cular, we show that the main property of smooth varieties (Theorem 1
of § 3) extends in a weaker version to normal varieties.
Theorem 2. If X is a normal variety, Y C X, and codim Y = 1, then
there exists an affine open set X' C X such that X' n Y i= 0 and the ideal
Y' = X' n Yin k [X'] is principal.
Of course, we may assume that X is affine. Let f E k [X], f i= 0,
fEoy. Then YC V(f), and since codim Y= 1 and codim V(f)= 1 (by
the theorem on the dimension of an intersection), Y consists of
components of V(f). Let V(f) = Yu y, Ycf. Y. Setting X = X - Y we
find that YnX i= 0, YnX = V(f)nX. Therefore we may assume at
once that Y= V(f).
To prove that Oy is a principal ideal means to fmd an element
u E Oy such that all the elements of Oy are divisible by u, that is,
OyU- l C k[X]. Such an element exists (possibly after replacing X by an
open set) if there is an element v E k(X) with the properties
oyvCk[X] , (3)
Oyvcf.Oy. (4)
For then there exists a u E Oy such that w = u . v ¢ Oy. Replacing X by
X - V(w) we achieve that w becomes invertible (in the ring k[X - V(w)]).
Since W¢Oy, we have Ycf.V(w) and (X-V(w))nYi=0. The element we
have found has the requisite two properties: u E Oy by construction
and OyU- 1 = OyVW- l = OyV c k [X - V(w)], because w is invertible in
k[X - V(w)].
Finally, (4) holds if (3) does and v ¢ k [X]. For ay has a finite
basis over k [X], and from the fact that OyV COy it follows that v is
integral over k [X]: this is one of the simplest properties of integral
elements. At this place we make use of the normality of X and conclude
that then v E k[X].
Thus, it is sufficient to construct an element v E k(X) such that
v¢k[X] and oyvCk[X]. We recall that Y= V(f). By Hilbert's Null-
stellensatz it follows that a~ C (f) for some I> 0, that is, the product of
any 1 factors (Xl' ••• ' (Xl E Oy is divisible by f We choose I as small as
possible, subject to this property. Then there exist (Xl' ••• ' (Xl-l E Oy
such that g = (Xl ••• (Xl- 1 ¢ (f), and g. (X E (f) for every (X E Oy, that is,
gOyc(f). So we see that we can set v=gf- l .
Theorem 3. The co dimension of the set of singular points of a normal
variety is not less than 2.
Let X be normal, dim X = n, S the set of singular points of X.
We have seen that S is closed in X. Suppose that S contains an
irreducible component Y of dimension n - 1. Let X' be the open set

whose existence was established in Theorem 2, and Y' = Yn X'. The
variety Y' has at least one simple point (as a point of Y', but not
necessarily as a point of X'). We denote it by y. Let (9y,y' be its local
ring on Y' and let u 1 , ..• , U n -1 be local parameters. By Theorem 2,
0Y' = (u), hence key'] = k[X']/(u). Similarly (9y, Y' = (9y,x'/(u). Evidently
my,X' is the inverse image of my,Y' under the natural homomorphism
(9y,X,~(9y.Y" We denote by v 1 , ... , vn - 1 arbitrary inverse images of
u 1, ... 'u n - 1. Then my ,x'=(v 1, ... ,Vn - 1,u). This shows that
dim my,x,/m;,X' ~ n, hence y is a simple point on X, against the
assumption that y E YC S. This proves the theorem.
Corollary. For algebraic curves the concepts of smoothness and of
normality are the same.
Let us make a comparison between the properties of normal
varieties we have deduced. First of all, observe that in the proof of
Theorem 1 we have not used the smoothness of X to the full extent,
but only the uniqueness of the decomposition into prime factors in
the rings (9x' In this context it is natural to single out the class of
varieties in which the latter property holds. They are called factorial.
Thus, a smooth variety is factorial, and a factorial variety is normal
(this is shown essentially by Theorem 1). It can be shown that all these
three classes of varieties are actually distinct. For example, it has been
proved that if a hypersurface in N, n;;" 5, has a unique singular
point, then it is factorial ([14], XI, 3.14). A pretty example of a factorial
surface that is not smooth is given by the equation x 2 + y3 + Z5 = O.
An example of a normal, but not factorial, variety is the quadric cone
we have already analysed: Z2 = (x + iy) . (x - iy), which has two distinct
decompositions of an element into prime factors.
Theorem 3 draws attention to a new property of varieties: the set
of singular points is of codimension not less than 2. Varieties with this
property are called non-singular in codimension 1. Theorem 3 asserts
that such are, in particular, the normal varieties. These two classes of
varieties are also distinct. We construct an example of a surface X that
is not normal, but has only finitely many singular points. To do this it
is enough to construct a regular finite mapping f: JA2 ~ JA4 such that
X = f(JA2) is closed in JA4, f: JA2 ~ X is a birational isomorphism, and
that two points, say Y1' YzEJA2 , have the same image ZEX, and that
f:JA2_{Y1'Y2}~X-{Z} is an isomorphism. Thus,fis very much like
the parametrization (2) of the curve (1) in Ch. I, § 1.1. The existence of
the mappingf contradicts the normality of X, and Z is the only singular
point on X. We specify fby the equation
f(x,y)=(x,xy,y(y-1), y2(y_1)).
If coordinates in JA..4 are denoted by u, v, w, t, then it easy to verify

that the equations of the variety X take the following form:
ut= VW, w3 = t(t- w), u 2w= v(v- u),
where u = x, v = xy, w = y(y - 1), t = y2(y - 1). The relations x = u,
y2_y=W show that x and yare integral overJ*k[X], hence thatJis
finite. The remaining properties of J we need are quite easy to verify.
2. Normalization of Affme Varieties. We consider the simplest ex-

ample of a non-normal variety, the curve X defmed by the equation
y2 = x 2 + x 3 . Its parametrization by means of t = y/x determines a
mappingJ: fA..1-+X or, what is the same, an embedding k[X] C k[t]. The
mappingJis a birational isomorphism, therefore k [X] C k[t] C k(X) = k(t).
The line JA..1 is normal, and consequently the polynomial ring k[t] is
integrally· closed. Furthermore, the ring k [t] can be characterized as
the collection of all elements u E k(X) that are integral relative to k [X].
For t 2 = 1 + x, hence t is integral over k[X], therefore all the elements
of k [t] are integral over k [X]. If u E k(X) is integral over k [X], then
it is also integral over k [t], and since k [t] is integrally closed,
u E k [t]. Finally, the fact that k [t] is integral over k [X] means in
geometrical terminology that the mapping J is finite. We show that for
any irreducible affine variety X there exists a variety X' and a mapping
X' -+ X with the same properties. We begin with a definition, which
refers to arbitrary irreducible varieties.
Defmition. A normalization oj an irreducible variety X is an
irreducible normal variety XV together with a regular mapping
v : XV -+ X that is finite and a birational isomorphism.
Theorem 4. An affine irreducible variety has a normalization that
is also affine.
Proof We denote by A the integral closure of k [X] in k(X), that is,
the collection of all elements u E k(X) that are integral over k [X]. From
the simplest properties of integral elements it follows that A is a ring
and integrally closed. Suppose that we have found an affine variety X'
such that A = k [Xl. Then X' is normal, and the inclusion k[X] C k[X']
determines a regular mapping J: X' -+ X. It is clear that X' is a
normalization of X.
According to Theorem 5 of Ch. I, § 3, such a variety X' exists if A
has no divisors of zero and if it is finitely generated. The first condition
is satisfied because A C k(X). The theorem will be proved if we can
show that A is finitely generated. We show even more, namely, that A is
finitely generated as a module over k[X].1f A = k[X] w l + ... + k[X]wm'
then W l , ... , W m , together with generators of the algebra k [X] over k
form a system of generators of A as a k-algebra. We use Theorem 10

of Ch. I, § 5. According to this theorem there exists a ring Be k [X]
over which k[X] is integral and which is isomorphic to a polynomial
ring: B~k[Tl' ... , T,.]. Let us draw a picture of all these rings and
fields:
BCk[X]CAC k(X)
" U
k(Tl' ... , T,.).
From this diagram and the simplest properties of integral elements it is

clear that A is the integral closure of B in k(X). Next, the field
K = k(X) is a finite extension of k(Tl , ... , T,.), because Tl , ... , T,. is a
transcendence basis of k(X). Finally, B is integrally closed (the variety
IN is normal and even smooth). Therefore the fmal result we need,
that A is finitely generated, is a consequence of the following
proposition.
Proposition.Let B=k[Tl , ... , T,.], L=k(Tl , ... , T,.), K a finite ex-
tension of L, A the integral closure of B in K. Then A is a B-module of
finite type.
The proof of the proposition differs depending on whether the
extension KjL is separable or not. Let us show how to reduce
everything to the case of a separable extension.
Let K = L (CXl' ... , cxs). If CXl is not separable over L, then its minimal
polynomial is of the form cx(m + a l cx«m-l) + ... + am = 0, where
ai E k (Tl' ... , T,.) and cx( is separable over L. We set ai = bf', where
bjEk(Tl'P', ... , T,.l/P}, L'=k(Tl'P·, ... , T,.l/P'), K'=K(Tl/p', ... , T,.l/P·);
B' = k[Tl/p', ... , T//P'], and let A' be the integral closure of B' in K'.
Then K' = L' (CXl' ..• , cxJ and exT + b l cxT- l + ... + bm = 0, so that CXl is
separable over L'. On the other hand, A C A', and if the proposition is
proved for A', then A' is a module of fmite type over B'. But B' is itself
a mOQule of finite type over B; a basis of it consists of the monomials
T{'/P', ... , T/r/ p', O~ il, ... , i,<p". Therefore A', and hence A, is a module
of fmite type over B.
So we see that the proof of the proposition reduces to the case
when CXI is separable. By the theorem on the primitive element there
exists an ex; EK such that L(ex l , ex 2 )=L(ex;). Then we have L(ex l , •.. , cxs )=
L(cx;, CX3' ... , CX.). Repeating the same arguments s-l times we reduce
the proof to the case of a separable extension.
For a proof in this case the reader is referred to [37], Vol. 1,
Ch. V, § 4, Theorem 7. This proof does not depend on the remaining part
of the book except for the rudiments of the theory of fmite extensions.
Theorem 5. If g : Y ~ X is a finite mapping and a birational iso-

morphism, then there exists a regular mapping h: Xv ~ Y such that the
diagram
commutes. Ifg: Y~X is a regular mapping, g(Y) is dense in X, and Yis

normal, then there exists a regular mapping h: Y ~ Xv such that the
diagram
Y
I~
Xv IX
commutes.
Proof of the first part. By assumption we have the embeddings
k [X] c k [Y] c k(X), where k [Y] is integral over k [X]. By the definition
of the integral closure k[Y]Ck[Xl, and this gives the required
regular mapping h : Xv~ y.
Proof of the second part. An element u of k[Xl is integral over
k [X] and is contained in k(X) C k(Y). Since k [Y]) k [X], a fortiori u
is integral over kEY], and because kEY] is integrally closed, we see
that u E k [Y]. Therefore k [Xl c k [Y], which gives a regular mapping
h: Y~XVwith the required properties.
Corollary. The normalization of an affine variety is unique. More
accurately, if v: Xv ~ X and v: Xv ~ X are two normalizations, then
there exists an isomorphism g : Xv ~ Xv such that the diagram
commutes.
This follows from either of the two parts of the theorem.
We do not prove the existence of a normalization for arbitrary
quasiprojective varieties. We mention that for those varieties for which
a normalization exists it has the property established in Theorem 5, as
follows immediately by considering affine coverings.
3. Ramification. The concept of normality allows us to derive one

important property of finite mappings. For a finite mapping f: X ~ Y
the nu~ber of inverse images of a point Y E Y is fmite, as we have

seen in Ch. I, § 5.3. Let us try to fmd this number. It is natural to
expect, by analogy with the theorem on the dimension of the inverse
image, that the number is one and the same for all points y in some
open set and that a variation can only arise on some closed subset
ZCY.
This is so in the simplest example of the mapping
(1)
To state the peculiarity of this example in general form we introduce

a concept.
Definition. If X and Yare irreducible varieties of equal dimension
and iff: X -+ Y is a regular mapping for which f(X) is dense in Y, then
the degree of the extension k(X)/ f*k(Y) (which under the present
conditions is finite) is called the degree of the mapping f:
degf= [k(X) :f*k(Y)] .
In the case of the mapping (1) deg f = 2, and if the characteristic

of k is not 2, then any point y i= 0 has two distinct inverse images,
whereas y = 0 has only one. Is the number of inverse images always
less than or equal to the degree of the mapping? This is not so in the
example of Ch. I, § 1.1, of the parametrization (2) f: Al -+ Yof the curve
Y with a double point (1); however, the inverse image of the singular
point consists of two points. It turns out that the reason for this is the
fact that the curve Y is not normal.
Theorem 6. If f: X -+ Y is a finite mapping of irreducible varieties
and Y is normal, then the number of inverse images of any point y E Y
does not exceed deg f
By virtue of the defmition of a fmite mapping we may confme our
attention to the case when X and Yare affine. We set
k[X]=A, keY] =B, k(X)=K, k(Y)=L, [K :L]=degf=n.
Since Y is normal, B is integrally closed, and since f is fmite, A is

a module of finite type over B. Therefore the coefficients of the
minimal polynomial of any element a E A lie in B. This is a simple
property of integrally closed rings, which the reader can find in [37],
Vol.t, Ch.V, §3. Let f-l(y)={Xl, ... ,xm }. We consider an element
a E A for which the values a(xj) are all distinct for i = 1, ... , m (if
X C/AN , then the matter comes to a construction of such a polynomial

in tv -dimensional space, and this is altogether elementary). Let
FEB[T] be the minimal polynomial of a. Clearly degF.,;;;n. We
replace in F all the coefficients by their values at y and we denote
the polynomial so obtained by F (T). It has m distinct roots a(xi ). Thus,
m .,;;; deg F= deg F .,;;; n ,
so that m .,;;; n, as required.
Henceforth we always consider [mite mappings f: X -+ Y of irre-
ducible varieties, where Y is assumed to be normal.
Definition. A mappingf is said to be unramified at a point y E Yif the
number of inverse images of this point is equal to the degree of the
mapping. Otherwise y is called a ramification point.
Theorem 7. The set of points at which a mapping is unramified is
open, and if the extension k(X)/f*k(y) is separable, it is not empty.
We keep to the notation introduced in the proof of Theorem 6.
Iff is unramified at y, then deg F= deg F = n, and F has n distinct roots.
We denote by D(F) the discriminant of the polynomial F. As we have
seen, the condition for being unramified at y can be written in the form
D(F)=D(F) (y)#O. (2)
But then D (F) (Y1 # 0 for the points y' of some neighbourhood of y.
This is what we had to show.
Thus, the set of ramification points is closed. It is called the
subvariety of ramification off
The question remains whether it is a proper subvariety. If the
extension k(X)/f*k(Y) is inseparable, then D(F)=O for the minimal
polynomial F of any element of this extension. Therefore the condition
(2) does not hold for even one point-all the points are ramification
points.
Suppose that the extension k(X)/f*k(Y) is separable. Thenfis also
called separable. Again we may take X and Yto be aifme, and we use the
previous notation. If a E A is a primitive element of the extension
k(X)/ f*k(Y) and F(T) its minimal polynomial, then deg F = n, D(F) # O.
Therefore there exist points y E Yat which D(F) (y) # 0, and hence f is
unramified. This proves Theorem 7.
We see that if a mapping f: X -+ Y is [mite and separable, the
varieties X and Yare irreducible, and Y is normal, then the picture is
the same as in example (1): the points of some non-empty subset U C Y
have deg f distinct inverse images, and the points of the complement
have fewer inverse images.
The following proposition expresses a fundamental property of unramified coverings:

Theorem 8.* Iff: X -+ Y is unramifled at ye Y, then for any point x ef-1(y) and any
r > 0 the homomorphism
qJ: f!J/m~-+f!JJm~, (3)
induced by the homomorphismf*: f!Jy-+f!J x is an isomorphism.

Intuitively this means thatfis "an isomorphism to within infinitesimals of any order".
The assertion of the theorem is local, therefore we may assume that X and Yare
affine and use the previous notation. For points x e X and ye Y we denote by ffix and
ffiy the maximal ideals ofthepointsx and y in the rings A and B, where A and Bare k[X]
and kEY] respectively. .
Consider the ring &= ~ 0 f!Jy and deri9te by mand m; its ideals generated by the ideals
~y C f!J y and ffi x, C A. It is easy to ~see that the rings f!J x, are isomorphic to the local rings
f!J m , of the maximal ideals mj of f!J. Therefore
f!J x.lm~, ~ &/mr
and the assertion of the theorem is equivalent to the fact that the natural homomorphisms
(4)
generated by the embedding f!Jy C & are isomorphisms.

The proof of this fact is based on two lemmas. In both we are concerned with ideals
a 1 , ... , an' a of some ring R, a l ) a, i = 1, ... , n. In this case the natural homomorphisms
R/a-+R/aj determine a homomorphism R/a-+ff)R/a j •
Lemma 1. The homomorphism

(5)
is an isomorphism.
The meaning of this homomorphism is that it associates with a function f e &
the collection of its values at the points XI' Hence it is clear that it is an epimorphism.
We need only prove that its kernel is trivial, that is,
(6)
Consider the ring l§ = &/m and its homomorphisms
whose kernels are the ideals mJm. We denote by u the image of an element u e & in l§.
As before, let a e A be an element for which all the a(xJ are distinct for i = 1, ... , n,
and let F(T) e f!Jy[TJ be its minimal polynomial. Going over to the residue classes
modulo mwe obtain an element i'iel§ and a polynomial Fek[T], with F(i'i)=O. Since
the Aii'i) = a(x j ) are all distinct and are roots of F, and since deg F = n, we have
F(T)= IT(T-Aii'i)).
i=l
Suppose that nmj#m. Then there exists an element uenmj such that u¢m.
By Hilbert's Nullstellensatz, US e m for some s> 0, and when we replace u by one of its
* Theorem 8, which is not as easy to prove as the remaining results of this chapter,
is used in the whole book only once: in the last section. Therefore the proof can be
omitted at a first reading.
powers, we may assume that U ¢ m, u2 Em. We apply to the element a + u the same
arguments as to a. Since Aj(ii) = 0, i = 1, ... , n, we see that a+u is a root of the same
polynomial F(T). Since u2 = 0, we have
(7)
From the fact that the polynomial F(T) does not have multiple roots it follows that
F(T) U(T)+ F(T) V(T)= 1
for suitable U, V E k [T]. Therefore F' (ii) U(ii) = 1, and from (7) it follows that u = 0,
UEm. This proves (6) and hence also Lemma 1.
Lemma 2. If at) a for a system of ideals a 1 , ••• , an, a of a ring R and if the
homomorphism
R/a ..... $ R/a j (8)
is an isomorphism, then the homomorphisms

R/ar ..... $R/ai
are isomorphisms for all r> O.
Lemma 2 follows from elementary properties of ideals, which we state without
proof. The reader can verify them or can consult [33], Vol. 2, § 89. The fact that the
homomorphism (8) is an isomorphism is equivalent to
a=naj, aj+aj=R, i#j.
From the second relation it follows that

ni+aj=R, i#j,
and nai = a'i ... a~. Hence
which proves Lemma 2.

Proof of Theorem 8. From Lemmas 1 and 2 it follows that
&/mr = $ mimi . ej (9)
where ej is the unit element of the ring mimi.

On the other hand, & is a module over the local ring {9,. We denote by U j elements
of & whose images are the ej. From Nakayama's lemma and (9) for r= 1 it follows that
the Uj, i= 1, ... , n, generate & over {9y. We show that they are even free generators of &
over (9,. This follows from the fact that if K and L are the fields of fractions of A and B,
respectively, [that is, K=k(X), L=k(Y)], then
K=&·L=Luj+ ... +Lu•.
Since [K: L] = n, U1' ••• , u. are linearly independent not only over {9y but even over L.
Thus, as a module over {9y
and hence
&/mr = {9y/m~ e 1 $ ... $ {9y/m~ e •.
Comparing this with the decomposition (9) we find that the homomorphisms (4) are
isomorphisms.
Corollary 1. Under the conditions of Theorem 8 the homomorphism of local rings

f* : {9y--> {9x determines an isomorphism of their completions1* :
@yC:::;@x'
The proof follows by a trivial verification from the defmition of a completion. The
assertion becomes perfectly obvious if we make use of the fact that @x is the projective
limit of the rings (9Jm~.
Corollary 2. Under the assumptions of Theorem 8 the tangent mapping
is an isomorphism.
This follows immediately from the theorem and the fact that
ex.x=(mx/m~)*, ey.x=(m/m;)*.
Corollary 3. Under the conditions of Theorem 8, X is smooth at x if and only if Y
is smooth at y.
This follows immediately from Corollary 2, the definition of a simple point, and the
fact that dim X = dim Y
In conclusion we explain the topological interpretation of the concept of being
unramified when X and Yare defined over the field of complex numbers. We assume
that the mapping f: X --> Y is finite, that X and Yare smooth, and that f is unramified at
all points of Y. In our case X and Y, as we have seen in § 2.3, are varieties of the same
dimension 2 dim X = 2 dim Y, and f is a continuous mapping. We show that f defines
X as an unramified covering of Y. This means that every point y E Yhas a neighbourhood U
with y E U such that f - I (U) splits into connected components Vi each of which is mapped
by fhomeomorphically onto U.
Letf-'(Y)={x" ... ,xm }, U"""U n be local parameters in a neighbourhood of y,
and V)i), ... , v~) be local parameters at Xi' Corollary 2 shows that [(iJvJiJu)(y)[ #0 for all
i = 1, ... , m. By the implicit function theorem it follows that there exist neighbourhoods
Vi of Xi and U of y such that f determines a homeomorphism between Vi and U for all i.
We can choose these neighbourhoods sufficiently small so that Vi and Vi are disjoint for
i # j. We show that f - , (U) = UVi, If y' E U, then f - '(yj consists of m points because
f is unramified. Since y' has m inverse images already in VI' ... , Vm , we have
f-'(U)=UVi·
4. Normalization of Curves
Theorem 9. A quasiprojective irreducible curve X has a normalization
Xv (which is also quasiprojective).
Proof Let X = UU i be a covering of X by affine open sets. We
denote by U; a normalization of Ui , which exists by Theorem 4, and by
/; : U/ ---> Ui the natural regular mapping, which is a birational iso-
morphism.
We embed the affine space containing U;v in a projective space and
denote by Vi the closure of Uiv in this projective space. Note that all the
varieties occurring so far are birationally isomorphic to X: Ui is
open in X, /; is a birational isomorphism between U/ and Ui , U/ is
open in Vi. Consequently U/ and J.j are birationally isomorphic. Let
({Jij : U/ ---> J.j be the corresponding mapping. According to the corollary
to Theorem 3, ut is a smooth curve, and since J.j is projective, ({Jij is
n
regular by the corollary to Theorem 3 of § 3. We set W= J.j, ({Ji = ({Jij'
j
n j
that is, q>i(U) = (q>i1 (u), q>du), ... , ... ). We denote by X' the union of all
the q>i(Un in W We claim that X' = XV. To substantiate this we have to
show that: a) X' is quasiprojective, b) X' is irreducible, c) X' is normal,
d) there exists a finite mapping v: X' ~ X that is a birational
n
isomorphism.
To prove all this we set Uo = Ui' which is an open subset of X.
From the construction of q>i it follows easily that U C o ur and that
all the q>i coincide on Ua. Let q> denote their restriction to Ua. Then
q>(UO)Cq>i(UnCq>(Uo), where q>(Uo) is the closure of q>(Uo) in W
Clearly q>(Uo) is an irreducible quasiprojective curve, and q>(Uo)-q>(Uo)
consists of finitely many points. By construction, q>(Uo) C X' C q>(Uo),
therefore q>(Uo) - X' consists of rmitely many points. This proves a)
and b).
Let x E X' then x E q>i(Un for some i, and q>i(Un is a neighbourhood
of x. We show that q>i is an isomorphism, and since Uiv is normal, it
then follows that X' is normal, which proves c). For this purpose we
observe that by construction q>ii is an isomorphic embedding of ut
in its closure lif. Therefore the mapping (u 1, U2' ... )~q>iil(Ui) is inverse
to q>i' which shows that it is an isomorphism.
Finally, to prove d) we construct a mapping· gi: q>i(Un~X;
gi = fiq>i- 1 . By the preceding all the gi are finite mappings. We show that
all the gi determine on X' a single finite mapping f : X' ~ X. To see this we
observe that all the gi coincide on q>i(UO) : if g: Uo~ Uo is a normalizing
mapping, then gi = 9 on UO. Therefore the mappings gi and gj agree
on the open subset q>(Uo), which is contained in q>i(Unnq>j(UJ). But
two regular mappings that agree on a non-empty open subset agree
everywhere; this follows from the corresponding property of functions.
Thus, gi and gj coincide at all points where they are both defmed, and
this means that all the gi determine a single regular mapping v : X' ~ X.
Clearly v is a birational isomorphism, and the theorem is proved.
Theorem 10. The normalization of a projective curve is projective.

Let X be a projective curve, XV its normalization, and v: XV ~ X
the normalizing mapping. We assume that the curve XV is not
projective and denote by X its closure in a projective space. Let
x EX - Xv, let U be some affine neighbourhood of x on X, U the V
normalization of U, and v': U ~ U the normalizing mapping. Then we

V
have the diagram

where <p and tp are isomorphic embeddings. The mapping V<p-1tpV'

is a birational isomorphism, and by Corollary 1 to Theorem 3 of § 3
and the fact that U V is a smooth curve, this mapping is regular. By
Theorem 5 the regular mapping h drawn on the diagram exists, and it
satisfies <ph = tpv'. However, its existence leads to a contradiction:
<ph(U") C Xv, and x E tpv'(U"), because the normalizing mapping is fmite
and hence epimorphic according to Theorem 4 of Ch. I, § 5. This
proves the theorem.
Corollary. An irreducible algebraic curve is birationally isomorphic
to a smooth projective curve.
This is a combination of the corollary to Theorem 3 and of
Theorem to.
The concept of normalization enables us to investigate properties
of curves in more detail.
Theorem 11. A regular mapping <p: X -+ Y is finite if X is an
irreducible smooth projective curve, dim Y> 0, and Y= <p(X).
Proof Let V be an affme neighbourhood of a point y E Y and
B=k[V]. We regard k(Y) as a subfield of k(X), the embedding being
effected by a mapping <p*. In particular, Be k(X); let A be the integral
closure of Bin k(X). In proving the existence of a normalization of an
affme variety we have explained that A is a ring of fmite type over B,
hence A = k [U], where U is an affme normal curve. Since it is
birationally isomorphic to X, by Corollary 2 to Theorem 2 of § 4 we
may assume that U is an open subset of X. We show that U=<p-1(V),
which guarantees the mapping is fmite.
Suppose that for some point Yo E V there is a point Xo f/: u,
<p(Xo) = Yo. We consider a function ff/: f9 xo ' fE (!JXt for all XiE U,
<p(xi) = Yo, Xi #= xo' Such a function is easy to construct by including the
points Xo and Xi each in an affme open set. If f has poles at points
x' E U, then <p (x1 = y' #= Yo, therefore we can fmd a function hE B
such that h(yo) #= O,.fh E (!Jx" that is, .fh E A. We have to take a function
that vanishes at the points y' and raise it to a sufficiently high power.
Now f1 =.fh is integral over B, that is,
Since f1 f/: (!J Xo' we have f11 E mxo' Therefore the last equation leads
to a contradiction: the right-hand side is regular at xo, but the left-
hand side is not.
Other applications are related to properties of singular points. In
fact, the existence of a normalization enables us to introduce some
useful characteristics of such points.
Let X be a curve and x a point of it, possibly singular; let v : Xv ~ X

be a normalization of X and Xl"", Xl inverse images of x on Xv.
The points Xi are called branches of X passing through x. This
terminology is explained by the fact that if k is the field of complex
(or of real) numbers and the Ui sufficiently small complex (or real)
neighbourhoods of the Xi' then some neighbourhood of x is the union
ofthe "branches" v(Uj).
We denote by 8 i the tangent space to Xv at Xi. The mapping dx, v carries
8 i into a linear subspace of the tangent space to X at x. Obviously
(dx, v) (8 i) is either the point x or a line. In the second case the branch
Xi is called linear, and the line (dXjv) (8 i ) the tangent to this branch.
The branch Xi is linear if and only if the mapping v* carries mJm;
into the whole space mx.lmi,. Suppose that x is the origin of coordinates
in the space JAn with the coordinates t l' ... , tn. Then
v*(t 1) + mi"
... , v*(tJ + mi,
generate v*(mx/m;). Since Xi is simple, dim mx.lmi, = 1, and there-
fore the branch Xi is linear if and only if v*(ts) ¢ mi, for at least
one s between 1 and n. To put it differently, v*(tJ must be a local
parameter at m x,. Since mx = (t1' ... , tJ, this condition of linearity in
invariant form becomes v*(mx) <t. m~,. As a measure of the deviation of a
branch Xi from linearity we can take the number I such that
v*(mx) C m~" v*(m x) <t. mk; 1. This number is called the multiplicity of
the branch Xi'
The point (0, 0) of the curve y2 = x 2 + x 3 gives an example of two
linear branches with the tangents y = x and y = - x, and the point
(0,0) on the semicubical parabola y2 = x 3 an example of a two-fold
non-linear branch.
If x is the centre of a unique linear branch, then x is a simple
point. This is a consequence of a lemma that will be proved in the
next subsection. Thus, the simplest characterization of the "singularity"
of a point is the number of branches corresponding to it and their
multiplicities.
A singular point of a plane algebraic curve is called ordinary
(or a point with distinct tangents) if only linear branches correspond
to it and if tangents to distinct branche.s are distinct.
5. Projective Embeddiogs of Smooth Varieties. The smooth pro-

jective model of an algebraic curve constructed in the preceding
subsection is situated in some projective space IPn. The question
arises how small this n can be chosen. We answer it by proving a
general result on varieties of arbitrary dimension.
Theorem 12. A smooth projective variety of dimension n is isomorphic

to a subvariety oflP 2n + 1. .
Let X be a smooth projective variety, Xc lPN. Theorem 12 will be
e
proved if we can choose for N> 2n + 1 a point E lPN - X such that
the projection from e is an isomorphic embedding of X in lP N - 1 •
Therefore we begin by clarifying when a regular mapping is an
isomorphic embedding.
Lemma. A finite mapping f of a smooth variety is an isomorphic
embedding if it is one-to-one and if dxf is an isomorphic embedding of
the tangent space ex for every x EX.
We set f (X) = y, <P =f - 1. The lemma will be proved if we can show
that <p is regular. This is an assertion of local character. Let y E Y
and f(x} = y, X E X. We denote by U and Vaffme neighbourhoods of
x and y such thatf(U)= V and that k[U] is integral over k[V]. The
restriction off to U is also denoted by f.
It is enough to show that f is an isomorphism for a suitable choice
of U and V. Then <p = f -1 is regular at y.
We recall that the space ex is dual to mJm;, where mx is the
maximal ideal of the local ring & x' The second condition of the lemma
means that the mapping f*: my/m; ~ mJm; is an epimorphism. In
other words, if my = (U1, ... ,u,), then f*(Ui} + m; generate mJm;.
Applying Nakayama's lemma (§ 2.1) to mx as a module over &x we
see that then mx = (f*(U1)' ... ,f*(1-l,}) or
( 1)
Now we apply Nakayama's lemma to &x as a module over f*(&y}.
e
Let us show that &x is of fmite type over f*(&y}. Let = Ct./P E &x,
that is, Ct., P E k[U], P(x} i= O. We denote by Z the set V(P}, which is
closed in U. From the Corollary to Theorem 4 in Ch. I, § 5, it
follows that T=f(Z) is closed. Since f is one-to-one, f- 1(y}=x,
therefore y ¢ T. Hence there exists a function bE k [V] such that
b=O on T, b(y}i=O. We setf*(b}=a. Then a=O on Z, a(x}i=O. By
Hilbert's Nullstellensatz there exists a natural number I and ayE k[U]
such that a'=Py. Hence '=Ct.y/a'Ek[U]f*(&y} and our assertion
follows from the fact that k[U] is of fmite type over k[V]. Now (1)
shows that &Jf*(my}&x=&Jmx=k, hence is generated by the
single element 1. From Nakayama's lemma it now follows that
&x= f*(&v}· Let U1' ... , urn be a basis of k[U] over keY]. By hypothesis,
Ui E &x = f*(&y}. We denote by V' = V - V(h} a smooth affine neighbour-
hood of y such that all the (f*) - 1(U;} are regular in U' = U - V(f*(h}).
Then k[U,]=~f*k[V']u;. By hypothesis, u;Ef*k[V'], from which it
follows that k[U'] = f*(k[V']}, and this means that f is an isomorphism
between U' and V'. The lemma is now proved.
Corollary 1. If every line pass{ng through a point ~ intersects X

in at most one point and if the tangent space to X at any of its points
does not contain ~, then the projection centered at ~ is an isomorphism.
It is sufficient to use Theorem 7 of Ch. I, § 5.
Now we can tum to the proof of Theorem 12. We need only show
that if X is a smooth variety, dim X = n, Xc IPN , N> 2n + 1, then
there is a point ~ satisfying the conditions of the corollary.
We denote by U1 and U2 the sets of points ~ E IPN relative to
which ~ fails to satisfy the first or second condition of the corollary,
respectively.
In IPN X X X X we consider the set r consisting of collinear
points (a, b, c), a E IP N, b, c E X. Clearly r is a closed subset of
IPN x X X X. The projections of IPN x X X X onto IPN and onto X x X
determine regular mappings ((J : r -t IPN and 1p : r -t X xX. It is evident
that if y E X xX, y = (b, c), b, c E X and in addition b =1= c, then
1p-1(y) consists of points (a, b, c), where a is any point of the line
passing through band c. Therefore dim 1p-1(y)= 1, and from Theorem 7
of Ch. I, § 6 it follows that dim r = 2n + 1. By defmition, U 1 = ((J(D,
and from the same theorem it follows that dim U1 ~ dim r = 2n + 1.
Similarly, to investigate the set U2 we consider in IPN x X the
set r consisting of those points (a, b) for which a E 8 b • In exactly the
same way we have mappings 1p:r-tx and ((J:r-tIP N. For XEX
we have dim 1p-1(X)= n, therefore dim r= 2n, and since U2 = ((J(D,
dim U2~2n.
So we see that dim U1 ~ 2n + 1, dim U2 ~ 2n, therefore, if N > 2n + 1,
then U1 u U2 =1= IP N , which is what we had to prove.
Corollary 2. Any quasiprojective smooth curve is isomorphic to a

curve situated in three-dimensional projective space.
We shall see later that not every curve is isomorphic to a curve
contained in a projective plane. Therefore not every algebraic curve
has a smooth plane projective model.
However, it can be shown that by extending the projection process
which we had used in proving Theorem 12 one can obtain a plane
curve whose singular points are all ordinary double points. According
to Theorem 12 every smooth surface is isomorphic to a surface
situated in a five-dimensional space. But generally speaking it cannot
be projected into four-dimensional space. However, the projection can
always be chosen so that apart from fmitely many points it is an
isomorphism. This leads naturally to examples of isolated points whose
local rings are not integrally closed such as the one constructed in § 5.1.
Exercises
1. Let X be an affine variety, and K a finite extension of k(X). Show that there
exist an affine variety Yand a mappingf: Y--+X having the following properties: 1)fis
fmite, 2) Y is normal, 3) k(Y)= K and f*: k(X)--+k(Y) determines the given embedding
of k(X) in K. Show also that Y is uniquely determined by these properties. It is called
the normalization of X in K.
2. Let X be the cone Z2 = xy. Show that the normalization of X in the field k(X) (Vx)
coincides with the affine plane and that the normalization mapping has the form
x=U 2 ,y=V2 ,z=uv.
3. Prove propositions similar to Exercise 1 for an arbitrary projective curve X. Show
that if X is projective, then so is Y.
4. How is the normalization of X x Y connected with the normalizations of
X and of Y?
5. Show that a point x is normal if the ring k [[X]] (see Exercise 14 of § 3) has no
divisors of zero and is normal. Hint: Use the result of Exercise 15 in § 3. Extend
Exercise 7 of § 3 to singular points and apply it.
6. Show that the cone X clAn given by the equation xi + ... + x; = 0 is normal for
n;;d.
7. Show that on the hypersurface X of Exercise 13 in § 3 the origin of coordinates
has an integrally closed local ring.
8. Is the Steiner surface normal? (Exercises 15 and 16 of § 1).
9. Show that every algebraic curve has a plane projective model in which the singular
points have linear branches only.
Chapter ill. Divisors and Differential Forms
§ 1. Divisors
1. Divisor of a Function. A polynomial in one variable is uniquely deter-
mined to within a constant factor by its roots and their multiplicities,
that is, by a collection of points Xl"'" Xr E /p,l with multiplicities
11 , ... , 1r · A rational function <p(x) = f(x)jg(x), 1, g E k[X] is determined
by the zeros of the polynomials f and g, that is, by the points at which it
vanishes or is non-regular. To distinguish the roots of g from those of f
we take their multiplicities with a minus sign. Thus, .•. , Xr with arbitrary integral multiplicities 11> ••• , Ir . Now we set
ourselves the task of specifying a rational function on an arbitrary
algebraic variety in a similar way.
We start out from the fact that according to the theorem on the
dimension of an intersection the set of points at which a regular
function vanishes forms a subvariety of co dimension 1. Therefore the
object we associate with a function is a collection of irreducible sub-
varieties of codimension 1 with preassigned multiplicities. These multi-
plicities can have positive and negative integer values.
DefInition. A collection of irreducible closed subvarieties C 1, .•. , Cr of
codimension 1 in an irreducible variety X with preassigned integral
multiplicities 11, ... , Ir is called a divisor.
A divisior D is written in the form
(1)
°
If all the Ii = 0, then we write D = 0, and if all the Ii> 0, then we write
D > and we say that D is effective. Irreducible subvarieties Ci of co-
dimension 1 taken with the coefficient 1 are called prime divisors. If in
(1) all the Ii # 0, then the variety C1 U ... U Cr is called the support of D
and is denoted by Supp D. If in (1) the prime divisors Ci are different,
we denote Ii by vc,(D).
Now we define an operation of addition of divisors. We observe that
if we allow the coefficients in (1) to take the value 0, any two divisors
can be written in the form
D'=I~Cl + ... +l;C" D"=I~Cl + ... +1~Cr
128 Chapter III. Divisors and Differential Forms
with the same C t , ... , Cr. Then, by definition,

Dr + D" = (l~ + ID C 1 + '" + (l~ + I;) Cr .
Thus, .the divisors on a variety X form a group isomorphic to a free
Z-module whose generators are irreducible subvarieties of co dimension
1 in X. This group is denoted by Div (X).
Now we proceed to assigning a divisor to a function 1 E k[X],
1 # 0. Let C be a prime divisor; first we associate with every function
1 E k(X), 1 # 0, an integer vdf). If X = fA1, this is the order of the zero
or pole of 1 at a point. .
This can be done only under a restriction on the variety X. Namely,
we assume that X is smooth in co dimension 1, that is, that the set of
singular points of X is of codjmension ~ 2. Let C C X be an irreducible
subvariety of codimension 1 and U an affine open set consisting of simple
points that intersects C and is such that C is determined in U by a local
equation. Such a set U exists under the restriction imposed on X, by
virtue of Theorem 1 in Ch. II, § 3. Thus, Qe = (n) in k[U]. We show that
for every function 1 E k[U], 1 # 0, there exists an integer I ~ 0, such
n °
that 1 E (n'), 1 ¢ (n ' + 1). If this were not the case that is, if 1 E (n') for all
I > 0, then 1 E (n'), therefore 1 = by Theorem 5 of Ch. II, § 2.
We denote the so defined number I by vdf). It has the properties
veU! 12) = vdIl) + vdI2), }

vdII + 12) ~ min(vdIl); vdI2)) (2)
for 11 + 12 #0
which follow easily from the definition and the irreducibility of C.
If X is irreducible, then any function 1 E k(X) can be represented in
°
the form 1 = g/h, g, hE k[U]. For 1 # we set vdf) = vdg) - vdh).
From (2) it follows at once that vdf) does not depend on the representa-
tion of 1 in the form g/h and that (2) holds for all 1 E k(X) other than
zero.
Our definition of the number vdf) so far depends on the choice
of the open set U, and we should write vgCf) instead of vdf). Let us
show that, in fact, vgCf) does not depend on U.
To begin with we assume that V is an affine open set, V C U, and
V n C # 0. Then n is a local equation of C also in V, and evidently,
v~Cf) = vgCf). But if V is any open set satisfying the same conditions
as U, then UnC and VnC are open in C and non-empty, and since C
is irreducible, their intersection is non-empty. Taking for W an affine
neighbourhood in Un V of some point x E Un V n C, we see that by the
preceding remark vgCf) = v:!'(f), v~Cf) = v:!'U), hence vgCf) = v~(f).
Thus, the notation vdI) is justified. Observe that if X = IA\ if C = x
§ 1. Divisors 129
is the point with the coordinate ct, and fEk[.d\l] = k[T], then vx(f) is
the multiplity of the root ct of the polynomial f(T), and the general
definition in essence copies this special case.
If vdf) = I> 0, we say that f has a zero of order I on C. If
vdf) = -I < 0, then f has a pole of order I on C. Observe that these
concepts are defined for subvarieties of co dimension 1 and not for points.
For example, for the function x/yon N the point (0,0) belongs both to
the subvariety of zeros (x = 0) and the subvariety of poles (y = 0) of the
function.
Now we show that to a given function f E k(X) there correspond
only finitely many irreducible subvarieties of co dimension 1 for which
vdf) =1= 0. First we consider the case when X is an affine variety and
f E k[X]. It follows from the definition that if C is not a component of
the subvariety V(f), then vdf) = 0. If X is affine, as before, but f E k(X),
then f = g/h, g, h E k[X], and we see that vdf)=O if C is not a component
of V(g) or V(h). Finally, in the general case, let X = U Vi be a finite
°
covering of X by affine open sets. Then every C intersects at least one Vi>
so that vdf) =1= only for those C that are closures of irreducible sub-
°
varieties CC Vi such that vc(f) =1= in Vi' Since the numbers of Vi and of
C in any Vi are finite, so is the number of C with vdf) =1= 0. Thus, we can
consider the divisor
Lvdf) C, (3)
where the sum extends over all irreducible subvarieties of co dimension 1
for which vdf) =1= 0. This is called the divisor of the function f and is
denoted by (f).
A divisor of the form D = (f), f E k(X), is called principal. If
(.f) = LliCi, then the divisors (.f)o = L liCi and (.f)oo = - L ljCj
i,I,>O j,lj<O
are called divisors of the zeros and of the poles of f Obviously,
eno)! 0, (f)ro )! 0, (f) = (f)o - (ftt:). We draw attention to some simple
properties:
Let us show that for a smooth irreducible variety X the converse is

also true: if (f) )! 0, then f is a regular function on X. Let x E X be a point
at whichfis non-regular. Thenf =g/h, h, gE{!)x while f¢{!)x' From the uni-
que prime factorization in {!)x (Theorem 2 of Ch. II, § 3) it follows that hand
g can be taken to be relatively prime in {!)x' Let n be a prime element of (!)x
that occurs in h but not in g. In some affine neighbourhood V of x the
variety V(n) is irreducible and of codimension 1. We denote its closure
in X by C. Then clearly vdf) < 0. This result is also true when X is a
normal variety, but we shall not prove this here.
Since on a projective irreducible variety X a function that is regular

at all points is a constant (Corollary 1 to Theorem 2 of Ch. I, § 5), from
the result we have just proved it follows that if (f);;;. 0, then
f = IX E k on a smooth projective variety X. In particular, on a smooth
projective irreducible variety a rational function is uniquely determined
by its divisor to within a constant factor: if (f) = (g), then (f. g- 1) = 0
and f = IX • g, IX E k.
Example 1. X =/A.n . By Theorem 3 of Ch. I, § 6, any irreducible
subvariety C of codimension 1 is given by a single equation: me = (F),
FE k[X]. Hence C = (F), that is, every prime divisor, and hence quite
generally every divisor, is principal.
Example 2. X = IPn. Every irreducible subvariety C of codimension 1
is given by a single homogeneous equation F, and Ue = (1;-1 F) in an
affine open set Ui if the degree of F is I. This leads to the following
method of constructing the divisor of a function f E k(IPn): we represent
f in the form f = F /G, where F and G are forms of the same degree,
and we decompose these forms into products of irreducible forms:
F=f1H!i, , G=f1L,,!j·
J ' then
if) = Iii Cj - ImjDj , (4)
where Cj and Dj are the irreducible divisors defined by the equations

Hj=O and Lj=O.
Let degF denote the degree of the form F. Since degF=degG,
we have ~/i degHi=~mjdegLj' We recall that in accordance with the
definition of the degree of a projective variety in Ch. I, § 6.5, deg Cj
is the same as degHi' Therefore in (4) ~/i degCi = ~mj degDj' Defining
the degree of the divisor D = ~/iCi as the number deg D = ~/i deg Ci,
we have shown that if D is a principal divisor, then degD = O.
It is easy to verify the converse: if ~/i deg Ci = 0 and Ci is given by
the equation Hi> where Hi is a form, then the function f = n Hli is
homogeneous of degree 0 and ~/iCi = if).
Example 3. The case X = IPn. x ... x IP n, is analysed similarly. Again
a subvariety C of codimension 1 is given by a single equation H = 0
(Theorem 3 of Ch. I, §6); however, H is homogeneous in each group of
coordinates of the spaces IPn, and accordingly has I distinct degrees
degjH(i = 1, ... , l). Just as in Example 2, we introduce the degrees
degiD of a divisor D on X, and D is principal if and only if degjD =0
(i=1, ... ,l).
The principal divisors form a subgroup P(X) of the group Div(X)
of all divisors. The factor group Div(X)/P(X) is called the group of
divisor classes and is denoted by CI(X). Divisors belonging to one
§ 1. Divisors 131
and the same coset in Div(X)/P(X) are called equivalent: Dl '" D2 if

D1 -D 2 =(f), fek(X). The cosets in Div(X)/P(X) are called divisor
classes. In the examples above we have
2. Locally Principal Divisors. We assume the variety X to be smooth.

In this case for every prime divisor C C X and every point x e X there
exists an open set U with x e U in which C is given by a local equation 'It.
If D is any divisor, D = }:,liCi, and if any of the Ci is given in U by the
local equation 'lti' then we have D = (f), f = II'It:'. Thus, every point x
has a neighbourhood in which D is a principal divisor. From all such
neighbourhoods we can choose a finite covering X = UU i, where in
every Ui we have D = (J;).
Evidently the functions J; cannot be chosen arbitrarily: the J; are
not identically zero, and in Ui n Uj the divisors (J;) and (ij) coincide. As
we have seen above, it follows that J; ij - 1 is a regular function in
Ui n Uj and does not vanish there. If a system offunctions {J;} correspond-
ing to the sets of the covering {Ui} satisfies the conditions that fJil is
regular and does not vanish in Ui n Uj, then we call it compatible.
Conversely, every compatible system of functions determines a
divisor on X. In fact, for every prime divisor C we set lc = vdfi) if
Ui n C "1= 0, where J; and C are regarded as a function and a prime divi-
sor in the variety Ui' From the compatibility of the system of functions
it follows that this number does not depend on the choice of Ui' Ob-
viously there are only finitely many C such that lc"l= 0, namely the
closures of the irreducible components of the divisors (J;). Therefore we
can consider the divisor D =}:, IcC. Obviously the given system of
functions {f;} corresponds to it.
Finally, it is easy to clarify when two systems of functions {J;} and
{gj} corresponding to coverings {UJ and {T'J}, respectively, give one
and the same divisor. For this it is necessary and sufficient that in
Uin T'J the functions .t;g; 1 are everywhere regular and do not vanish.
The simple verification is left to the reader.
The specification of divisors by systems of functions enables us to
investigate their behaviour under regular mappings. Let cp: X --+ Y be a
regular mapping of smooth irreducible varieties, and let D be a divisor on
y. We assume that cp(X) 1- SuppD. We show that under this restriction
we can determine the inverse image cp*(D) of D by analogy with the deter-
mination of the inverse image of a regular function. First of all we
clarify when the inverse image of a rational function f on Y can be
constructed, and when it does not vanish identically on X. For this it is
sufficient that there exists at least one point ye cp(X) at which f is
regular and fey) =1= o. Then such points form a non-empty open set V.
Now f is regular on V, and hence cp*(f) is a regular function on
cp -leV) that does not vanish identically (in fact, nowhere). Since cp - I (V)
is open in X, we see that cp*(f) determines a rational function on X.
In terms of divisors our condition on the mapping cp and the function f
reduces to the fact that cp(X) ct Supp(f).
Suppose now that the divisor D is given by a compatible system of
functions {fi} and a covering {VJ. We consider those Vi for which
cp(X)n Vi is not empty, and we show that cp(X)n Vi ct Supp(jJ For it
follows from the irreducibility of X that cp(X) is irreducible in Y. If
cp(X)n Vie SuppCt;), then it follows from the irreducibility of cp(X) and
the fact that cp(X)n Vi is non-empty, that cp(X) C Supp(jJ Finally, the
facts that Suppct;) n Vi = Supp D n Vi' that cp(X) is irreducible, and that
it intersects Vi imply that cp(X) C Supp D, against the assumption.
Hence for all Vi that intersect cp(X) the rational functions cp*(j;) are
defined in cp - 1 (VJ The sets cp - 1 (V;) = V; for which cp(X) intersects Vi
are open and form a covering of X, and the functions cp*(D form a
compatible system, which determines some divisor on X. Obviously
this divisor does not change when D is given by another system of
functions. The divisor so obtained is called the inverse image of D and
is denoted by cp*(D).
In particular, if cp(X) is dense in Y, then the inverse image of any
divisor DE Div(Y) is defined.
If D and D' are two divisors on Y given by systems of functions of
{lJ and {gjl, corresponding to coverings {V;} and {Vj}, then the divisor
D + D' is given by the system of functions {j;. gj} and the covering
{Vi n Vj}. From this it follows at once that cp*(D + D') = cp*(D) + cp*(D'),
so that if cp(X) is dense in Y, then cp* defines a homomorphism
cp* : Div Y~ Div X .
The principal divisor (f) is given by the system of functions
/; = f, consequently cp*((f») = (cp*(f») .
Therefore cp* maps P(Y) into P(X) and defines a homomorphism
cp* : Cl(Y)~CI(X).
As an application of the specification of divisors by compatible
systems of functions we show how to associate a divisor not with a
function, but with a form in the coordinates on a smooth projective
variety. Let Xc IPN and let F be a form in the coordinates in IPN that does
not vanish identically on X. For every point x E X we consider a form G
of the same degree as F such that G(x) =1= O. Such forms exist: for
example, if x = (IXo: ... : IXN) and IX; =1= 0, we can take G = T1 egF• Then
f = FIG is a rational function on X and is regular in the open set in
which G =1= O.
§ 1. Divisors 133
It is easy to see that there exist forms G; such that the open sets
U; = X - X Gi form a covering of X. It is just as easy to verify that the
functions /; = F/G; and the open subsets U; form a compatible system
of functions and hence determine a divisor on X. Another choice of
forms G; does not change this divisor, which therefore depends only the
form F. It is called the divisor of F and is denoted by (F). Since the
}; are regular in the sets U;, we have (F) ~ O. If Fl is another form,
degFl =degF, then (F) -(F1) is the divisor of the rational function
F/F1 • Therefore (F) ",(F1 ) if degF =degF1 .
In particular, all divisors (L), where L is a linear form, are equi-
valent to each other. Evidently Supp(L) = XL' the section of X by the
hyperplane L = O. Therefore they are called divisors of a hyperplane
section.
Taking above for F1 the form Ldeg F we obtain that (F) '" degF· (L),
where (L) is the divisor of a hyperplane section.
All the arguments connected with the specification of a divisor
of a compatible system of functions can be generalized to arbitrary,
not necessarily smooth, varieties. However, here the possibility of speci-
fication by a compatible system of functions must be taken as the defini-
tion of a divisor. The object at which we arrive in this way is called a
locally principal divisor.
Strictly speaking, a locally principal divisor on an irreducible variety
is a system of rational functions {f;} corresponding to the open sets of a
covering {U;} and satisfying the conditions:
1) the}; do not vanish identically and
2) fJj - 1 and fj/;-1 are regular on U;n Uj •
Here two sets of functions {/;} and {gj} and coverings {U;} and {Vj},
respectively, determine the same divisor if /;gj 1 and };-1 gj are regular
in U;n lj.
Every function f E k(X) determines a locally principal divisor by
setting}; = f Such divisors are called principal.
The product of two locally principal divisors given by functions
{/;} and {gJ and coverings {U;} and {Vj}, respectively, is the divisor given
by the functions {f;gj} and the covering {U;n VJ. All locally principal
divisors form a group, and the principal divisors a subgroup of it. The
factor group is called the Picard group of the variety X and is denoted by
Pic(X).
Every locally principal divisor has a support: this is the closed
subvariety consisting of those points in U; at which }; is non-regular
or zero. Just as for divisors on smooth varieties, so we can define the
inverse image of a locally principal divisor D on Y under a regular
mapping q>: X -+ Y if q>(X) is not cqntained in SuppD.
We mention one important special case. If X is a smooth variety and

Ya subvariety, not necessarily smooth, then any divisor D on X for which
Supp D 1> Y determines a locally principal divisor fj on Y. To see this we
have to consider the embedding cp : Y -+ X and to set fj = cp*(D). We call
fj the restriction of D to Y and denote it by Qr(D). From the definition
it follows that for principal divisors Qy{(f)) = (j), where j is the restric-
tion of ! to y.
Of course, the distinction between divisors and locally principal
divisors appears only in the case of non-smooth varieties.
3. How to Shift the Support of a Divisor Away from Points

Theorem 1. For every divisor D on a smooth variety X and finitely many
points x 1> ... , xm E X there exists a divisor D' such that D' ,..., D, Xi ¢ Supp D'
(i= 1, ... , m).
We can take D to be a prime divisor, because otherwise it would be
enough to apply the theorem to each of its comp!tlnents. In X we choose
an affine open set containing the points Xl' ... , X m • It is sufficient to prove
the theorem for this set, so that we may assume X to be an affine
variety. Using induction on m we may assume that Xl' ... , Xi ¢ SuppD,
Xi+ 1 E SuppD. It remains to construct a divisor D' such that D',..., D,
Xl' ... , Xi+ 1 ¢ SuppD'. We consider some local equation re' of the prime
divisor D in a neighbourhood of Xi + l' Let us show that re' can be
chosen so that re'E k[X] (by assumption, X is affine). Indeed, re' is regular
at X i + I' and hence, if (re')oo = 'Lkli, then x i + 1 ¢ Fl' Hence for every
I there exists a function fz E k[X] vanishing on Fi and such that
fz(x i + d # O. Evidently the function re = re' rrH' is regular on X and is a
local equation of D in a neighbourhood of Xi+I' Since by hypothesis
Xj¢SUPpDUXIU",UXj_IUXj+IU",UXiU=1, ... ,i), for every
j = 1, ... , i there exists a function 9j E k [X] such that 9jlD = 0, 9 j(x l ) = 0
(I = t. .... j - t,j + t, ... , i), gix) #0.
We consider the function
i
!=re+ L (XjgJ,
j=l
(XjEk,
and choose the constants (Xj so that

!(x j )#OU=1, ... ,i). (1)
It is sufficient to take (Xj # -re(xj)/gix/. Since all the gjlD = 0, in the
local ring (9 Xi+ 1we have g/=: O(re) and 'L(XjgJ = re 2h, hE(9xi+ 1'/ = re(1 + reh).
Since (1 + reh) (x i + I) = t, it follows that! is a local equation of D in a
neighbourhood of X i + I' Therefore (f) = D + 'L rsDs' and none of the
prime divisors of Ds passes through Xi + l' This means that if we set
§ 1. Divisors 135
D' = D - (f), then Xi + 1 ¢ Supp D'. Furthermore, (1) shows that

Xj ¢ Supp(f) (j = 1, ... , i), therefore the divisor D' satisfies the conditions
of the theorem.
Here is a first application of Theorem 1. In § 1.2 we have defined
the inverse image f*(D) of a divisor D of a variety X under a regular
mapping f: Y --+ X, provided that f(Y)it SuppD. Theorem 1 enables
us to replace D by an equivalent divisor D' for which X ¢ SuppD',
where x is an arbitrarily chosen point in f(Y). Then automatically
f(Y) 1:. SuppD', and the inverse image f*(D') is defined. This shows
that without any restrictions on the regular mapping f we can define
the inverse image of a divisor class C E CI(X). For this purpose we have
to choose in C a divisor D such that f(Y) 1:. SuppD and consider the
class on Y containing the divisor f*(D). It is easy to verify that in this
way we obtain a homomorphism
f*: CJ(X)--+CI(Y).
In other words, CI(X) is a functor from the category of irreducible

smooth algebraic varieties into the category of Abelian groups.
4. Divisors and Rational Mappings. The correspondence between

functions and divisors is useful for the investigation of rational mappings
of varieties into a projective space. Let X be a smooth variety and
cp: X --+ IPn a rational mapping. We wish to find out at what points cp
is non-regular.
A rational mapping is given by formulae
(1)
where we may assume that none of the functions J; vanishes identically

on X. Let
m
(J;) = L kijCj ,
j= 1
where the Cj are prime divisors. Here we allow some k ij to be zero.

To clarify whether cp is regular at a point x E X we specify Cj by a
local equation nj at x. Then
h =(9 nJ'i) Ui' U i E (!)x, ui(x) #0.
By the unique prime factorization in (9x there exists a greatest

common divisor d of the elements fo, ... , J", that is, an element dE k(X)
such that hd-l E (9x and if dlE k(X) is such that J;d1 l E (9x, then dlld,
that is, dd 11 E (!)x.
Since local equations of irreducible varieties are prime elements

of (f) x' we have
k j = min k. ..
i=O •... ,n lJ
The mapping cp is regular at x if there exists a function g E k(X),

such that J;g-l E (f) x (i = 0, ... , n), and the (J;g -1) (x) are not all zero. By
the definition of the greatest common divisor it follows that g Id. If
d = g. h, hE (f)x, and h(x) = 0, then hl(J;g-l), hence all the (J;g-l) (x) =0.
Thus, only a function g for which d = g . h, h(x) # 0, can satisfy the
necessary conditions. Then J;g-l = (J;d- 1) h, that is,
J;g-l =(ry n~ij-kj)(Uih),
and cp is a regular mapping if and only if not all the functions nJirkJ n
vanish at x. j
To express this result in the language of divisors we define the g.c.d.

of divisors Di =l:.kijCj (i = 1, ... , n) as the divisor
g.c.d.(D1' ... ,Dn )= l:.kjCj , kj =. min kij.
1=1, ···,n
Obviously Di = D; - g.c.d.(D1' ... , Dn) ~ 0, and the divisors D; have no

common components. In particular, let us set
D = g.c.d.((fo), ... , (fn)) , D; = (J;) - D.
Then in some neighbourhood of x
(ry n~ij-kJ) = D; ,
and we can say that the mapping cp is regular at x if and only if not all
the varieties SuppD; pass through this point.
So we have proved the following result.
Theorem 2. The rational mapping (1) is non-regular precisely at the
points of the set
n SuppD;, D; =(J;) - g.c.d·(fo), ... , (fn)) (i =0, ... , n).
Since the divisors D; do not have common irreducible components,
the set n SuppD; is of codimension ;:. 2. Thus, Theorem 2 is a sharper
form of Theorem 3 in Ch. II, § 3.
Remark. The divisors D; can be interpreted as the inverse images of
the hyperplanes Xi=O under the mapping cp:X ~IPn. For if x¢nSuppD;
and D = (h) in a neighbourhood of x, then in the same neighbourhood
a regular mapping is given by the formulae:
cp = (fo/h : ... :fjh).
§ 1. Divisors 137
The inverse image of the hyperplane Xi = 0 has the local equation fJh,
hence coincides with D;.
More generally, if ..1.= (..1.0: ... : An) and E). C]pn is the hyperplane
L AiX i = 0, then
5. The Space Associated with a Divisor. The fact that all polynomials
f(t) of degree ,;;;;; n form a finite-dimensional vector space can be
interpreted in terms of divisors in the following way. We denote by Xoo
the point at infinity on the projective line ]pI with the coordinate t.
A polynomial in t of degree I has a pole of order I at Xoo and has no
other poles. Therefore the condition degf,;;;;; n can be expressed as
follows: the divisor (f) + nxoo is effective.
By analogy, for an arbitrary divisor D on a smooth variety X we
can consider the set consisting of zero and of those functions
f E k(X), f i= 0, for which
(1)
This is a linear space over k under the usual operations on functions.
For if D = L niCi, then (1) is equivalent to the fact that
ve;(f) > - ni , vdf) > 0 for C i= Ci
and by virtue of this our assertion follows immediately from the
formulae in § 1.1.
The space of functions satisfying the conditions (1) is called the
space associated with the divisor D and is denoted by 2(D).
Just as polynomials of degree ,;;;;; n form a finite-dimensional space,
so the space 2(D) is finite-dimensional if D is an arbitrary divisor and
X a projective variety.
In § 2 this theorem will be proved for the case of algebraic curves.
By an induction on the dimension it can then be proved in the general
case without any particular difficulty. However, the place of this
theorem becomes more intelligible if it is obtained as a special case of
a vastly more general proposition on coherent sheaves. In this form
it will be proved in Ch. VI, § 3.
The dimension of the space 2(D) is also called the dimension of the
divisor D and is denoted by I(D).
Theorem 3. Equivalent divisors have equal dimensions.
Let DI '" D2 ; this means that DI - D2 = (g), g E k(X). If f E 2(DI)'
then (f) + DI > O. From this it follows that (f. g) + D2 = f + DI > 0,
that is, f· g E 2(D2)' g. 2(D 1) = 2(D 2 ). Thus, multiplication of all
functions .f E 2(D I ) by a function g determines an isomorphism of the
spaces 2(D I ) and 2(D2)' and the theorem follows.
So we see that we can talk of the dimension l( C) of a divisor

class C, understanding by this the common dimension of all the divisors
in this class. This number has the following meaning. If DEC,
IE2(D), then the divisor Df=(.f)+D is effective. Clearly Df~D,
therefore Df E C. Conversely, every effective divisor D' E C is of the
form D f' where IE 2(D). Obviously, if X is projective, then the
function I is determined by the divisor Df uniquely to within a constant
factor. Thus, we 'can set up a one-to-one correspondence between the
effective divisors of the class C and the points of the (l( C) - 1)-
dimensional projective space lP(2(D)) corresponding to D. (We recall
that the projective space lP(L) corresponding to a vector space L
consists of all lines of L).
The space 2(D) is useful in specifying rational mappings by
divisors, as this was described in § 1.4. If
cp = (fo : ... : In) : X ~ lP n (2)
is a rational mapping and, as in § 1.4,

(3)
then Di ~ 0, hence all the /; E 2( - D).
The choice of the functions .r; depends on the chosen system of
projective coordinates in lPn. Therefore, to the mapping cp there
corresponds, in an invariant fashion, the totality of all the functions
n
L: Ai/; that are linear combinations of the /;. These functions form a
i=O
linear subspace Me 2( - D). In what follows we assume that cp(X) is
not contained in any proper linear subspace of lPn. Then L Ai/; =1= 0 on
X if not all the Ai = O. The set of effective divisors corresponding to
this set of functions, that is, the divisors (g) - D, gEM, is called a linear
system of divisors. If M = 2( - D), then the linear system is called
complete. The meaning of the divisors (f) - D, IE M, is very simple:
they are the inverse images of the divisors of the hyperplanes in IPn
under cpo Thus, we can construct all rational mappings of a given smooth
variety X into various projective spaces. For this purpose we must take
an arbitrary divisor D, and in the space 2( - D) a linear finite-
dimensional subspace M. If Io, .. .,fn is a basis of it, then the formula
(2) gives the required mapping. Observe that the divisors Di for these
IiE2(-D) have an additional property: they have no common com-
ponents.
Since multiplication of all functions .r; by a common factor g E k(X)
does not change the mapping cp, but a divisor D is changed into an
equivalent divisor (g) + D, the class of the divisor D is an invariant of
§ 1. Divisors 139
the rational mapping. Thus, we have the following method of con-

structing all those rational mappings cp of a variety X into a projective
space ]pm for which cp(X) is not contained in any proper subspace of ]pm:
we choose an arbitrary divisor class on X, and for every divisor D of
this class we choose in 2( - D) a linear finite-dimensional subspace M
such that the effective divisors (f) - D have no common components.
If io, ...,.r..
is a basis of M, then our mapping is given by (2). Of course,
it can happen that 2(-D)=0 or that all the divisors (f)-D,
i E 2( - D) have a common component; then this divisor class does not
lead to such a mapping.
We draw attention to one interesting property of this situation.
Among all rational mappings corresponding to a given class C there
exists a maximal one: this is obtained by taking for M the whole space
2( - D), DEC. (Here we rely on the unproved theorem that the space
2( - D) is finite-dimensional.)
All other mappings corresponding to this class are obtained by
constructing the compositum of this mapping with various projection
mappings. For if cp = (fo : ... : iN) and, say, 1p = (fo : ... : in)' n < N, then
1p=ncp, where n(xo: ... : x N ) =(xo: ... : xn) is a projection, which we now
regard as a rational mapping.
Let us see how this scheme works if we take for X the projective
space ]pm. We know that CI(]pm)::::z and that the class CI corresponding
to an integer I consists of the divisors of degree I.
Clearly, if I> 0, D E C/, then 2 (- D) = O. If I ~ 0, then we can take
for - D the divisor - IE, where E is the divisor of the hyperplane at
infinity Xo = O. In this case 2( -IE) consists of polynomials of degree
~ -I in inhomogeneous coordinates xtfx o, ... , x,Jxo (see Exercise 15).
By multiplying the resulting formulae for the mapping by x~, we
obtain a Veronese mapping VI: ]Pm-+]P'"m. So we see that every
rational mapping of ]pm can be obtained by combining a Veronese
mapping with a projection.
Exercises
1. Determine the divisor of the function x/yon the quadric xy-zt=O in lP 3 .
2. Determine the divisor of the function x -1 on the circle xi + x~ = x~, x = x,/xo.
3. Determine the inverse image f*(D a ), where fix, y) =X is the projection of the circle
of Xl + y2 = 1 onto the x-axis, and Do is the divisor on the line A', Da = p, pEA', with
the coordinate a.
4. X is a smooth projective curve, f E k(X). Regarding f as a regular mapping
f: X .... lP', prove that (f) = f*(D), where D = 0 - CIJ is a divisor on lP'.
5. X is a smooth affine variety. Show that Cl(X) = 0 if and only if factorization in
k[X] is unique.
6. X is a smooth projective variety, X C lPN, k[S] is the ring of polynomials in
inhomogeneous coordinates in lPN, and Wx C k[S] is an ideal of X. Prove that if in the ring
k[S]/IRx factorization is unique, then CI(X) = 71., and a generator is the class of hyperplane
sections.
7. Find CI(lP" x /A").
8. The projection p: X x /A1 ..... X determines a homomorphism p*: CI(X) ..... CI(X x JAI).
Show that p* is an epimorphism. Hint: Use the mapping q*: CI(X x JAI) ..... CI(X), where
q: x ..... X X JAI is given by q(x) = x x o. .
9. Show that for every divisor on X x JAI there exists an open set U C X such that
on U x JA! this divisor is principaL Hint: X can be regarded as affine, and the divisor as
irreducible. Then it is given by a prime ideal in k[X x JAI] = k[X] [T]. Use the fact that
in k(X) [T] all ideals are principal, then replace X by some principal affine open subset.
10. Show that CI(X x JAI) "" CI(X). Use the results of Exercises 8 and 9.
11. Let X be the projective curve given by the -equation y2 = x 2 + x 3 in affine
coordinates. Show that every locally principal divisor X is equivalent to a divisor whose
support does not contain the point (0, 0). Use this and the normalization mapping
tp:lPI ..... X for which tp-I(O,O) consists of two points Xl and X2ElPI to describe Pic(X)
as D/P, where D is the group of all divisors on lPI whose supports do not contain Xl and
X2, and P is the group of those principal divisors (f) for which! is regular at Xl and X 2
and !(x I ) = !(x 2 ) # o. Show that Pic (X) is isomorphic to the multiplicative group of
non-zero elements of k.
12. Find Pic(X), where X is the curve with the equation y2 =X3.
13. Let X be a quadric cone. Use the mapping tp : JA2 ..... X described in Exercise 2 to
Ch. II, § 5, to determine the image tp*(Div(X)) in Div JA2 . Show that D = (F) E Div JA2
belongs to tp*(Div(X)) if and only if F( - u, - v) = ± F(u, ·v), that is, F is either an even or
an odd function. Show that principal divisors on X correspond to even functions. Show
also that CI(X) "" 71./271..
14. Using Theorem 2 determine the points at which the birational mapping
tp: X ..... lP 2 is non-regular, where X is a quadric in lP 3 , and tp the projection from a point
X E X. The same for tp-l.
15. Show that if E is the hyperplane Xo = 0 in lP", then the space ff' (IE) consists of the
polynomials in inhomogeneous coordinates xtfx o, ... , xjxo of degree.;;; I. Hint: Use the
fact that if! E ff' (IE), then! E k[JA~].
16. Show that every automorphism of lP" carries divisors of hyperplanes into each
other. Hint: The class of hyperplanes is determined by invariant properties in CI(lP"),
and the divisors of hyperplanes as effective divisors in it.
17. Show that every automorphism of the variety lP" is a projective transformation.
Hint: Use the result of Exercise 16.
18. Let (J: X ..... Y be the (J-process centred at y E Y, where Y is smooth. Show that
CI(X) "" CI(Y)EB71..
§ 2. Divisors on Curves
1. The Degree of a Divisor on a Curve. We consider a projective smooth
curve X. A divisor on X is a linear combination of points D = "2:.k i x i ,
ki E 7/." Xi E X. The degree of the divisor D is defined as the number
degD = "2:.k i ·
Example 2 of § 1.1 for n = 1 shows that on X = IPl a divisor D is
principal if and only if deg D = O. We now show that deg D = 0 for a
principal divisor on any smooth projective curve. For this purpose we
make use of the concept of the degree of a mapping f, degf, which was
introduced in Ch. II, § 5.3.
Theorem 1. If f: X --+ Y is a regular mapping of smooth projective

curves and f (X) = Y, then degf = degf*(y) for every point ye Y.
In Theorem 1 !*(Y) is the divisor on X that is the inverse image
of the divisor on Y consisting of the point y with the coefficient 1. Thus,
degf is equal to the number of inverse images of any point y e Y (taken
with appropriate multiplicities). This makes the intuitive meaning of
the degree of f clearer-it shows how many times X covers Yunder f.
Corollary. The degree of a principal divisor on a smooth projective
curve X is equal to zero.
F or every non-constant function f e k(X) determines a regular
mappingf:X--+IPl. Heref*(O)=(f)o for OeIPl-this follows at once
from the definition of the two divisors. Similarly f*( 00) = (f)oo' By
Theorem 1,
deg(f) = deg(f)o - deg(f)oo = degf*(O) - degf*( 00) = degf - degf = O.
If X and Yare two varieties of the same dimension and if f is a
regular mapping f: X --+ Y such that f (X) is dense in Y, then it deter-
mines an embedding f*: k(Y)--+k(X); utilizing this we shall henceforth
regard keY) as a subfield of k(X) ( that is, for u( key) we write u instead
of f*(u) when this cannot lead to misunderstandings}.
Theorem 1 follows from two results. To state them we introduce
the following notation. Let Xl' ... , Xr be points on the curve X. We set
&= n
i= 1. ...• r
(!)x,· (1)
Thus, & consists of the functions that are regular at all the points
Xl' ... , X r • If {Xl' ... , X r } = f-l(y), ye Y, then the ring (!)Y' which_we have
agreed above to regard as a subring of k(X), is contained in (!).
Theorem 2. & is a principal ideal ring with finitely many prime ideals.
There exist elements t; e &such that
Vx,(t) = ~ij' 1 ..;; i , j..;; r . (2)

Ifue&, then
(3)
where Ii = vx,(u), and v is invertible in &.
~eorem 3. If {Xl> ... , x r } = f-l(y), then &is a free module over (!)y
and (!) ~ (!);, where n = degf.
Let us first show how Theorem 1 follows from Theorems 2 and 3.
Let t be a local parameter at y, and {x 1, ..• , x r } = f-l(y). According to
Theorem 2, t = t~1 ... r;v, where Ii = vx,(t). Recalling the definition of
the inverse image of a divisor we see that

r
f*(y) = I:.lix i and degf*(y) = L Ii·
i=1
Since the elements t 1, ... , tr are pairwise coprime in ifi, we have

r
ifiI(t) ~ EB ifiI(tl') .
i= 1
It is easy to see that every element W E ifi has a unique representation

in the form
W == 0(0 + 0( 1 ti + ... + 0(1, -1 tl' - I (mod t;') , O(i E k . (4)
For if we have already got the representation
W == 0(0 + 0(1 ti + ... + O(s_1tf-1 (modtf),
then
v = t i- S (W - 0(0 - ••• - O(s _ 1 tf - 1) E ifi c (fJ x, .
We set v(x i) = O(S. Then v.~,(v - 0(,) > 0, and from Theorem 2 it follows
that v == O(s (mod til, that is,
W == 0(0 + 0( 1 ti + ... + 0(,-1 [,-1 + O(stf (mod tf + 1).
This proves (4) by induction. _
From the representation (4) it follows that dimifiI(tl') = Ii. Therefore
r
dimifil(t)= L Ii· (5)

i= 1
When we now apply Theorem 3, it follows that ifil(t) ~((fJy/(t))". But t is

a local parameter at y, therefore
(fJy/(t) ~ k, dimifij(t) = n = degf. (6)
Now (5) and (6) prove Theorem 1.
Proof of Theorem 2. We denote by U i a local parameter at Xi. Then
Xi occurs in the divisor (u i) with the coefficient 1, that is, (u;) = Xi + D,
where Xi does not occur in D. By Theorem 1 of § 1 we can shift the
support of D away from X 1, ... , X r ' that is, we can find a function /; such
that these points do not occur in D + (/;). This means that for ti = ui /;
the relations (2) hold. Let U E ifi. We set vx.(u) = Ii. By hypothesis,
Ii;;;' 0. For the element v=utllt~··t;lr we ha'ye VXI(V) =0 for all i= 1, ... , r,
from which it follows that v E (fJ and V-I E (fJ. So we obtain a representa-
tion (3) for u. _-
It remains to verify that (fJ is a principal ideal ring. Let a be an ideal of
ifi. We set Ii = inf UEa
vx(u)
l
and a = til ... t~r. Then ua- 1 E ifi, that is, a C (a).
.§ 2. Divisors on Curves 143
Let us show that a = (a). To do this we denote by a' the set of functions
ua-l, UEa. Evidently a' is an ideal of 19 and infvx'(u)=O. Hence for
uea' 1
every i = 1, ... , r there exists a Ui E a' for which Vx;(Ui) = 0, that is,
ui(x;):#:O. An obvious verification shows that vxlc) =0 (i=1, ... ,r) for
r
the element c= L u j t 1... tj... tr Ea'
j=l
(the symbol tj indicates that the
_
corresponding factor is absent). This means that c- 1 E (!), therefore
a' = 19, a = (a). This proves the theorem.
N ow we turn to the proof of Theorem 3. First of all, we show that
19 is a module of finite type over (!}y. For this purpose we recall that
according to Theorem 11 of Ch. II, § 5, the mapping f is finite. There-
fore the point y has an affine neighbourhood V such that the curve
U = f -1 (V) is also affine and that the ring A = k[U] is a module of
finite type over B = kEY]' As always, the embedding Be A is effected by
the mappingf*.
Lemma. In the previous notation 19 = A(!}y, even if Y is not a normal

curve.
For if cp E 19 and Zi are the poles of cp on U, then f(Zi) = Yi ¥- y. There
exists a function hE B such that hey) ¥- 0, h(Yi) = 0, and cph E (!}z;' hence
ceh E A. Since h- 1 E (!}y, we have cp E A(!}y' So we have shown that
(!) e A(!}y- The reverse inclusion is obvious, and the lemma is proved.
Now we can complete the proof of Theorem 3. Clearly, generators
of the module A o_ver kEY] are at the same time generators of A(!}y
over (!}y. Therefore (!) is a module of finite type. By the main theorem on
modules over a principal ideal ring, 19 is a direct sum of a free module
and a torsion module. However, (!}y and 19 are contained in the field
k(X), from which it follows that this torsion module is zero and that
19 ~ (!);' for some m. _
It remains to determine m, that is, the rank of @. It is equal to the
ma~imum number of linearly independent elements over @y contained
in (!). Since linear independence over a ring and over its field of fractions
is one and the same thing, and since the field of fractions of @yisk(Y),
we see that m is equal to the maximal number of linearly independent
elements of 19 over k( Y).
By hypothesis, [k(X): key)] = n, so that necessarily m.,;; n. It remains
to show that 19 contains n linearly independent elements relative to
key). Let (Xl' ••• , (Xn be a basis of the extension k(X)jk(Y). We denote by
I the maximum order of the poles of the functions (Xi at the points Xj'
and by t a local parameter of y. Evidently the functions (Xitl are regular
at these points, hence are contained in i9. Consequently, they are linearly
independent over kEY]' This completes the proof of the theorem.
2. Bezout's Theorem on Curves. Here we give the simplest applica-

tions of the theorem on the degree of a principal divisor. They are very
special cases of more general theorems, which we shall prove in connec-
tion with the theory of intersection indices. However, it is convenient
to give an account of these simple cases now, because they will be useful
for us in the next subsection.
Let X be a smooth projective curve, X C lPn, F a form in the point
coordinates oflP n that is not identically zero on X and x a point on X.
In § 1.2 we have introduced the divisor (F) of F on X. The degree
deg(F) of this divisor is also denoted by (X, F) and is called the inter-
section index of X with the hypersurface lP;.
Theorem 1 leads at once to an important consequence: this number
is one and the same for all forms of the same degree.
For if degF = degFl' then f = FIFI E k(X). From the definition of
the divisor (F) it follows at once that (F) = (F1 ) + (f), hence (F) '" (Fl)' By
the corollary to Theorem 1, deg(F) = deg(Fd.
To find out how the number (X, F) depends on the degree of the
form F it is sufficient to take for F any form of degree m = degF. In
particular, we may set F = Lm , where L is a linear form. Then
(X, F) = m(X, L) = (degF) (X, L). (1)
Finally, we explain the meaning of the number (X, L). In Ch. I we have
introduced the concept of the degree degX of a curve X as the
maximum number of points of intersection of X with a hyperplane not
containing X.
Since (X, L) = L vx(L), we have degX ~ (X, L).
L(x)=O
Let us find out when vx(F) = 1 for the case of an arbitrary form F.
By virtue of the additivity of the function vx(F) it is sufficient to consider
the case of an irreducible form.
Lemma. Let X C lPn, F an irreducible form, and Y = lP;. The equality
vA(F») = 1 is equivalent to the fact that F(x) = 0, and ex,y 7J ex,x' Both
these spaces are regarded as subspaces of e.~,lPn.
The proof comes from a comparison of some defmitions in Ch. II.
Let G be a form for which G(x) # 0, deg G = degF. By definition,
vx(F) = vx(f), where f = (F IG)lx. We know that vx(f) > 1 is equi-
valent to the fact that f E m~. or, what is the same, dxf = O. But
dxf E e~,x is also the restriction to e X,x of the differential dx(FIG) of
the function FIG, which is rational on lP n and regular at x. Thus,
vAF) > 1 is equivalent to dAFIG)=O on ex,x' Furthermore, FIG is a
local equation of Y in a neighbourhood of x in which G # O. Therefore
dx(FIG) =0 is the equation of e d and dAFIG) =0 on ex,x if and only
if eX,y) e x.x '
We apply this to calculate the intersection index (X, L).

Since the number (X, L) is one and the same for all linear forms L,
the number of points x E X for which L(x) = 0 assumes its maximum
when all the vAL) = 1. By the lemma this is equivalent to the fact that
the hyperplane L does not touch X at any point. Taking for L such a
linear form we find that
degX = (X, L) . (2)
It only remains to verify that linear forms with the required
property actually exist. This is easily done by means of arguments we
have used many times: in the product X x pn (where pn is the space of
hyperplanes in ]pn) we consider the set r of pairs (x, ~) such that ~ touches
X at x. A standard application of the theorem on the dimension of
fibres of mappings then shows that the image of r under the projection
X x pn --+ pn is of codimension ~ t.
Comparing (1) and (2) we obtain the relation
(X, F) = degF . degX , (3)
which is called Bezout's theorem. This theorem has many applications
in elementary geometry, which one can find, for example, in [34],
Ch. III.
3. Cubic Curves. From the corollary to Theorem t it follows that all

equivalent divisors on a smooth projective curve have the same degree.
Hence we can speak of the degree of a divisor class. We have there-
fore the homomorphism
deg: CI(X)--+Z
whose image is the whole group Z and whose kernel consists of the
classes of degree zero and is denoted by ClO(X). The role of this group
will already be clear from the following result.
Theorem 4. A smooth projective curve X is rational if and only if
ClO(X) =0.
For if X ~]pl, we are concerned with Example 2 of § 1.1 (for n = t).
There we have seen that CI(JP1)=Z and hence ClO(JP 1) =0. Conversely,
let ClO(X) =0. This means that every divisor of degree zero is principal.
In particular, if x, y E X, x 1= y, then there exists a function f E k(X)
such that x - y = (f). Regarding f as a mapping X --+]pl we deduce
from Theorem 1 that k(X) = k(f), that is, f is a birational isomorphism.
Since X and ]pi are smooth projective curves, f is an isomorphism.
Now we analyse the simplest case when ClO(X) 1=0. These are plane
smooth projective curves of degree 3. In Ch. I, § t we have seen such
curves need not be rational; for example, the curve with the equation
x 3 + y3 = 1 is non-rational. In § 5.4 we show that all plane smooth pro-

jective curves of degree 3 are non-rational. We shall now make use of
this fact.
Theorem 5. We choose an arbitrary point Xo on a smooth projective
plane curve X of degree 3 and associate with any point x E X the class C x
containing the divisor x - Xo' The mapping x --+ Cx determines a one-to-
one correspondence between points x E X and classes C E ClO(X).
If C x = C y , x - Xo ~ Y - Xo and x ~ y. From the proof of Theorem 4
it follows that for every x i= y this would lead to the curve X being
rational, whereas we know that it is not.
It remains to show that in every class C of degree zero there is a divisor
of the form x - Xo' To begin with, let D be any effective divisor. We
show that there exists a point x E X such that
D~x+lxo· (1)
If degD=I, then (1) is true with 1=0. If degD>I, then D=D'+y,
degD' = degD - 1, D' > O. Applying induction we may assume that (1)
is proved for D': D' ~ z + mxo. Then D '" y + z + mxo. If we can find a
point :x such that
y+z~x+xo , (2)
then (1) follows.
First let y i= z. We draw the line through these points with the
equation L = O. By Bezout's theorem (L, X) = 3, and hence
(L) = Y + z + u, U EX. (3)
Next we suppose that U i= xo, and we draw the line through U and
Xo with the equation Ll = O. As in (3) we find that (L 1 ) = u + Xo + x.
Since (L) ~ (L 1), we have y + z + u ~ u + x + x o, hence (2) follows.
We still have to analyse the cases when y = z or u = Xo. If y = z,
then we draw the tangent to X at y. Let L = 0 be its equation. By the
Lemma in § 2, vy((L));;;, 2, and therefore (L) = 2y + u. Thus, (2) also holds
in this case. The case u = Xo is treated similarly.
Now let degD =0. Then D = D J -D2,Dl;;;' 0, D z ;;;' 0, degDI =degD z.
Applying (1) to Dl and D z we see that Dl '" Y + lxo, D z ~ z + lxo with
one and the same I, because degDI = degD z. Therefore
D=D 1 -D z "'y-z,
and it is sufficient to find a point x for which y - z '" x - Xo. This is
equivalent to y+ Xo '" z+x and is the same as (2) apart from the notation.
4. The Dimension of a Divisor. In § 1.5 we have associated with a di-

visor D on a smooth variety a vector space 2'(D).
Theorem 6. The space 2'(D) is finite-dimensional for every divisor
D on a smooth projective algebraic curve.
First of all it is easy to reduce the assertion of the theorem to the case
D ~ O. For let D = Dl - D2, DJ ~ 0, D2 ~ O. Then ll'(D)C ll'(Dd: if
fEll'(D), then (f)+D 1 -D2 =D' ~O, hence (f)+D 1 =D' + D2 ~O, that
is f E ll'(D1). This gives the required reduction. Now let D ~ 0,
r
D= L nixi, ni ~ O. At the points Xi we choose local parameters t i.
i= 1
The condition f E ll'(D) is equivalent to VXi(f) ~ - ni(i = 1, ... , r),
vx(f) ~ 0 for X =I- Xi' that is f E tini(!)x, (i = 1, ... , r), f E (!)x for X =I- Xi·
In view of all this we can consider the linear mapping
r
q>: ll' (D) -+ EB ti ni(!)x/(!)Xi
i=1
that associates with a function fEll' (D) all its residue classes in the spaces
tini(!)x/(!)x,. If q>(j)=0, then fE(!)x, 0= 1, ... ,r), and since fEll'(D),
we have f E (!)x for X =I- Xi. Therefore f is regular at all the points
X E X. Since X is a projective curve, such a function must be a
constant Thus, the kernel of q> is k, hence one-dimensional. To show
that ll'(D) is finite-dimensional it remains to verify that the space
r
EB ti-n,(!)x'/(!)x,
is finite-dimensional. Obviously multiplication by ti'
i=1
determines an isomorphism ti n, (!)x'/(!)x, '4 (!)x,/ti' (!)x" and in the proof of
Theorem 2 we have seen that the space (!)x,/ti.' (!)x, is of finite dimension
EB t i- n,(!)x'/(!)x,
r
ni. Thus, is a direct sum of finite-dimensional spaces,
i= 1
hence itself finite-dimensional.
Together with the proof of the theorem we have obtained the estimate
dimll'(D) ~ degD + 1 for D ~ o.
Exercises
1. Let X be a smooth affine curve, and Xl' ..• ,XmEX. Show that the functions ti in
Theorem 2 can be taken to be the left-hand sides of the equations of those hyperplanes E,
for which X, E E" Xj ¢ E, for i i-- j and E, 7J ex"x (so that it does not touch X at Xi).
2. Show that if a curve X is non-rational, then the estimate I(D)..;; degD + 1 in
Theorem 6 can be improved to I(D)..;; degD for every D ~ o.
3. Let X be the projective closure of the affine curve y2 = X3 + Ax + B, where the
polynomial X3 + Ax + B does not have multiple roots and the characteristic of k is
different from 2. Show that X is a smooth curve and that its intersection with the line at
infinity consists of a single point Xo. Find a local parameter at Xo and the numbers
vxo(x), vxo(y)·
4. Under the conditions of Exercise 3, find the general form of a function in ft' (mxo).
In particular, show that l(mxo) = m for m > O. Utilize the fact that every function f E k(X)
can be written in the form P(x) + Q(x) y, P, Q E k(X), and find out when f E ft'(mxo).
5. Under the conditions of Exercise 3 find out how to express addition of classes
in Clo in terms of the points corresponding to them, according to TheoreIl! 5. More
accurately, if Xl' X2 E X, Cx • = Cx ! + CX2 ' find out how to express the coordinates of X3 in
terms of coordinates of Xl and X2' Here Xo can be taken to be the point at infinity on X.
6. In the notation of Exercise 3 show that Cx ! + Cx • + Cx • = 0 if and only if the points
Xl> X2, and X3 are collinear.
7. In the notation of Exercise 3 show that if X = (IX, p), - Cx = Cy , then y = (IX, - Pl.
Show that the group Clo has exactly four elements of order two. Find the points on X
corresponding to them.
8. A simple point X E X, where X is a plane curve, is called a point of inflexion if
vx (8 x • x ));;' 3. Show that under the conditions of Exercises 3, 4, and 5 X is a point of
inflexion if and only if 3 Cx = O.
9. Show that the line passing through two points of inflexion of the curve X of
Exercises 3-7 intersects it in a third point of inflexion.
§ 3. Algebraic Groups
The results of the preceding sections lead to an interesting branch of
algebraic geometry: the theory of algebraic groups. We do not go
deeply into this topic, but to give the reader at least some idea of it, we
give in this section an account of some of its main results, omitting most
of the proofs.
1. Addition of Points on a Plane Cubic Curve. Theorem 5 of §2 esta-
blishes a one-to-one correspondence between the points of a smooth
projective plane cubic curve X and the elements of the group cqX).
To a point x E X there corresponds the class Cx containing the divisor
x - Xo, where Xo is a fixed point, which serves to specify the cor-
respondence.
Making use of this we can transfer the group law from ClO(X) to
the set X itself. The resulting operation on points of X is called addition
and is denoted by EEl. According to the definition, x EEl y = z if Cx + Cy = Cz ,
that is,
(1 )
Evidently the point Xo is the null element. We denote it henceforth by 0,
so that (1) can be rewritten in the form
x + Y - (x EEl y) + 0 . (2)
The proof of Theorem 5 of § 2 makes it possible for us to describe
the operation EEl and the operation e of taking the opposite element
in elementary geometric terms. Namely, if the tangent to X at 0 intersects
X at p, and if the line passing through p and x intersects X at x: then
20+ p- p+x+x', x +x' ....,20 (3)
e
which means that x' = x (Fig. 7). If x = p, then the line through x we
have drawn must be replaced by the tangent at p..
§ 3. Algebraic Groups 149
Similarly, to describe the operation EB we draw the line through

x and y. Let z' be its third point of intersection with X and z the third
point of intersection with X of the line passing through z' and o. Then
(Fig. 8).
x+y+z'",z'+z+o,
x+y",z+o, (4)
z=xEBy.
If x = y (or z' = 0), the secant through x and y must be replaced by
the tangent at x (or at z').
Now we prove the important property of "algebraicity" of the group
law on X. It will be the basis of the definition of an algebraic group in
the next subsection.
Theorem 1. The mappings cp: X ...... X, cp(x) = ex, and 1p: X x X ...... X,
1p(x, y) = xEB y, are regular.
Lemma. Let a E X and let Sa: X ...... X be the mapping that associates
with a point x the third point of intersection of L with X, where L is the
line passing through x and a If x =1= a, and the tangent at a if x = a.
Then the mapping sa is regular.
We begin by showing that Sa is regular at all points x =1= a. For this
purpose we choose a coordinate system so that a is the origin of the co-
ordinates and that the third point of intersection with X of the line L
joining x to a lies in the finite part of the plane. The latter condition
is easily seen to mean that if x = (e, 11) and f(u, v) = f3(U, v) + f2(U, v)
+ f1 (u, v) is the equation of X, then f3(e, 11) =1= O. The equation of L is of
the form u = te, v = t11. Substituting this in the equation of X we obtain
the equation t 3f3(~' 1]) + t 2f2(~' 1'f} + t fl (~, 17) = O. We know two of its
roots t = 0 and t = t corresponding to the points of intersection a and x.
Therefore the value t corresponding to the third point of intersection is
obtained from the relation 1 + t = - f2/ f3' So we see that
~x
( ) = (_ f3(~' 1]) + f2(~.1]) ~,
;: _ f3(~' 1]) + f2(~' 1]) )
1].
f3(~,1]) f3(~,1])
Since f3(~' 1]) # 0, this shows that Su is regular for x # a.
To prove the same for x = a we observe that a rational mapping of
a smooth projective curve into itself is regular. Hence there exists a
s;
regular mapping Sa that coincides with Sa for x # a. Since = 1, Sa and
Sa are one-to-one. But as they agree at all points except possibly one,
there must also agree at this point. Hence Sa = Sa' and this means that Sa
is regular.
Proof of Theorem 1. The assertion about q> follows immediately from
the lemma, because according to (3), q> = sp. The mapping 1p(x, y) for
x # y is defined by means of the secant through x and y. Arguing as in the
proof of the lemma, we can easily verify that 1p is a rational mapping.
For any point a E X the mapping ta, tu(x) = aEf) x, is regular, because
according to (4) ta = So Su' Obviously the relation
1p(x, y) = t;;~b 1p(tu(x), tb(Y»)
holds for arbitrary points a, bE X. Therefore, if 1p is regular at
(xo, Yo), then it is also regular at (ta(xo), tb(YO»)' But it must be regular at
some points, because it is rational. Hence it follows that it is regular
everywhere.
2. Algebraic Groups. Plane cubic curves are one of the most impor-
tant examples of a general concept, which we now introduce.
An algebraic group is an algebraic variety G which at the same
time is a group for which the following conditions hold: the mappings
q>:G-G, q>(g) = g-1, and 1p:G x G-G, 1p(g1' g2)= 9192 are regular (here
9- 1 and 9192 are the inverse elements and the product in G).
Examples of Algebraic Groups
Example 1. A plane cubic curve with the group law Ef). The fact that
the conditions in the definition of an algebraic group are satisfied is the
contents of Theorem 1.
Example 2. The affine line /AI on which the group law is given by ad-
dition of point coordinates. This group is called additive.
Example 3. The variety ./AI - 0, where 0 is the origin; the group law
is given by multiplication of point coordinates. This group is called
multiplicative.
Example 4. In the space N 2 of square matrices of order n the open

set of non-singular matrices with the usual law of matrix multiplication.
This is called the general linear group.
Example 5. In the space N 2 the closed subset consisting of the ortho-
gonal matrices. Naturally, the group law is the same as in Example 4.
Let us show by a very simple example how the fact that G is an al-
gebraic group can influence the geometry of the variety G.
Theorem 2. The variety Qf an algebraic group is smooth.
From the definition of an algebraic group it follows that for any
hE G the mapping
is an automorphism of G.
Since th(gt)=g2 for any g1' g2EG where h=g2g1 \ and since the prop-
erty of a point of being singular is invariant under automorphisms, we see
that if at least one point of G is singular, then so are all the points.
But this contradicts the fact that in any algebraic variety the singular
points form a closed proper subvariety. Therefore G cannot have
singular points.
3. Factor Groups. Chevalley's Theorem. This subsection contains the

statement of some basic theorems on algebraic groups. Proofs of these
theorems are not provided.
A subgroup of an algebraic group G is a subgroup of G that is a
closed subset of G.
A subgroup He G is called normal, as in the abstract theory of
groups, if g - 1 H g = H for all g E H. Finally, a homomorphism of al-
gebraic groups q;: G 1 -* G2 is a regular mapping that is a homomorphism
of abstract groups.
The problem of constructing the factor group of a given normal
subgroup N is very delicate. The difficult question is, of course, how to
turn the set GIN into an algebraic variety.
TheoremA.*The abstract group GIN can be made into an algebraic
group in such a way that the following conditions hold:
1. The natural mapping q;: G-*GIN is a homomorphism Qf algebraic
groups.
2. For any homomorphism of algebraic groups 1p: G-*G 1 whose
kernel contains N there exists a homomorphism f: GIN -*G 1 for which
1p =.r q;.
* Letters denote theorems that are stated without proof.
Obviously the algebraic group GIN is uniquely determined by the

conditions 1 and 2. It is called the factor group of N in G.
An algebraic group G is called affine if the algebraic variety G is
affine, and it is called an Abelian variety if the algebraic variety G is
projective and irreducible.
Theorem B. An affine algebraic group is isomorphic to a subgroup of a
general linear group (Example 4 above).
Evidently the general linear group, and hence each of its subgroups,
is affine.
Theorem C (Chevalley's Theorem). Every algebraic group G has a
normal subgroup N such that N is affine and GIN is an Abelian variety.
N is uniquely determined by the these properties.
4. Abelian Varieties. The condition of projectivity of the variety of an

algebraic group G, which defines Abelian varieties, contains a surpris-
ing amount of information. Many unexpected properties of algebraic
varieties follow from it. We derive the simplest of these here, because
they only require appliations of simple theorems that were proved in
Ch. I.
We need a property of arbitrary projective varieties. We define a
family of mappings of the variety X into Z as a regular mapping
f :X x Y-+ Z, where Y is some algebraic variety, the so-called base of
the family.
Evidently, for every y E Y we have the mapping fix) = f(x, y),
which justifies our terminology.
Lemma. ~r X and Yare irreducible varieties, X is projective, and,
for a family f of mappings of X into Z with base Yand some point Yo E Y,
f(X x Yo) is a single point Zo E Z, then f(X x y) is a single point for
every yE Y
Proof Consider the graph r of f Obviously re X x Y x Z and r
is isomorphic to X x Y We denote by p the projection X x Y x Z
-+ Y x Z, and by f the set p(T). Since X is projective, f is closed by
Theorem 3 of Ch. I, § 5. We denote by q: f -+ Y the mapping defined
by the projection Y x Z-+ Y The fibre of q over y obviously has the form
(y,f(x, y», hence is not empty, so that q(f) = Y. On the other hand, by
hypothesis, for y = Yo the fibre consists of the single point (Yo, zo). Ap-
plying Theorem 7 of Ch. I, § 6, we see that the fibres over an open set
are zero-dimensional and that dimf = dim Y
We take an arbitrary point Xo E X; clearly f) {(y,f(xo, y», y E Y}.
Since both varieties are irreducible and have the same dimension,
they are identical, and this means that f{X x y) = f(x o, y).
(u=o, v=o)
Fig. 9
Note. Without the assumption that X is projective the lemma is

false, as is shown by the example of the family of mappings
f:JA? x ./AI --+JA\ f(x, y) =X y. The reason for this is that the set r is
not closed arid that Theorem 7 of Ch. I, § 6 is not applicable to it.
r
In our example C./AI x./A1 =./A2 consists of all the points (u, v) except
those with u = 0, v i= O. This is a" plane from which the line u = 0 has been
removed but the point u = 0, v = 0 retained. Actually, Theorem 7 of
Ch. I, § 6 is not true for the projection q: (u, v)--+u: the dimension of
the fibre over the point u =0 is 0, but the dimension of the image is 1,
and the dimension of the variety to be mapped is 2 (Fig. 9).
Theorem 3. An Abelian variety is commutative.
Consider the family of mappings of G into G with basis G:
f:G x G--+G,j(g, g')= g-I g' g.
Evidently f(g, e) = e, hence by the lemma f(G, g) consists of a single
point. Therefore f(G, g')=f(e,g')= g', but this means that the group G
is commutative.
Theorem 4. If 1p : G --+ H is a regular mapping of an Abelian variety
G into an algebraic group H, then 1p(g) = 1p(e) cp(g), where e EGis the unit
element and cp : G --+ H is a homomorphism.
Proof We set cp(g) = 1p(e)-I1p(g) and show that cp is a homomorphism.
For this purpose we consider the family of mappings of the variety G
into H whose base coincides with G:
f:G x G--+H,f(g',g)=cp(g')cp(g)cp(g'g)-I.
Since cp(e) = e' is the unit element of H, we have f( G, e) = e'. By the lemma,
the image f( G, g) for every element g E G consists of a single point, so that
f(g', g) does not depend on g'. Settingg' =ewe see thatf(g', g)= f(e, g)=e',
which means that cp is a homomorphism.
Corollary. If two Abelian varieties are isomorphic as algebraic varie-
ties, then they are also isomorphic as groups - "the geometry determines
the algebra".
5. Picard Varieties. The only examples of Abelian varieties which
we have encountered so far are plane cubic curves. We have defined a
group law on them, starting out from their group of divisor classes.
This example is typical for a far more general situation. Starting out
from arbitrary smooth projective variety X we can construct an Abelian
variety whose group of points is isomorphic to a subgroup of CI(X)
(or ClO(X) in the case of a cubic curve). We give this definition, but
omit the proofs of all assertions except the very simplest. Our aim is
to study divisors on smooth varieties, but in the course of the arguments
we come across divisors on arbitrary varieties. In that case we under-
stand by divisors only locally principal divisors.
We now define a new relationship of equivalence for divisors: algebraic
equivalence. It is coarser than (that is, follows from) the equivalence
we have considered before.
Let X and T be any two irreducible varieties. For every t E T the
mappingjt: x~ (x, t) defines an embedding of X in X x T. Every divisor
C on X x T for which Supp C"jJ X x t determines a divisor j1'( C) on X.
In that case we say that the divisor.it (C) is defined.
A family of divisors on X with base T is a mapping f: T ~ Div(X).
A family f is called algebraic if there exists a divisor C E Div(X x T)
such that the divisor ji(C) is defined for all tE T andji(C) = f(t).
Two divisors Dl and D2 on X are called algebraically equivalent
if there exists an algebraic family of divisors f on X with base T and
two points t 1, t2 E T such that f(t d = DI, f(t 2) = D2. This relation is
written as DI == D2. Thus, algebraic equivalence of two divisors indicates
that it is possible to "deform them algebraically" into each other.
Clearly algebraic equivalence is reflexive and symmetric. It is easy to
show that it is also transitive. If the algebraic equivalence of two divisors
DI and D2 is realized by a divisor C on X x T and the equivalence of
D2 and D3 by a divisor C' on X x T, then to show that Dl and D3 are
equivalent we have to consider the divisor
(C x T) + (C' x T) - D2 X TxT
on X x TxT. The detailed verification is left to the reader.
with addition in the group Div(X): the divisors D with D == form a

subgroup, which we denote by Div"(X).
°
Finally, it is easy to see that algebraic equivalence is compatible
Equivalence of divisors implies their algebraic equivalence. It is

enough to verify this for equivalence of a divisor to zero. Let
DE Div(X), D ~ 0, that is, D = (g), g E k(X). Consider the variety
T =;p,2 - (0, 0) and denote by u and v coordinates on ;P,2. We regard
g, u· and v as functions on X x T, understanding by this, as usual,
p*(g), q*(u), and q*(v), where p: X x T ~ X and q: X x T ~ T are the
projections. We set C = (u + vg) and consider the algebraic family deter-
mined by the divisors C on X x T. It is easy to verify that f( 1,0) = 0

(the divisor zero), f(O, 1) = D, hence D == O.
Finally, we consider the notion of algebraic equivalence on the ex-
ample of a smooth projective curve X. For any two points x, y E X
we have x == y. To see this it is sufficient to consider the family of
divisors f parameterized by X itself and defined by the diagonal on
X x X. It is easy to verify that f(x) = x for all x E X. Therefore D == (~ni)xO
for every divisor D = ~ nix i and every point Xo E X, that is, two divisors
of the same degree are algebraically equivalent.
The converse is somewhat more difficult to prove: algebraically
equivalent divisors on a smooth projective curve have the same degree.
We do not prove this here. Thus, divisors on a smooth projective curve
X are algebraically equivalent if and only if they have the same degree.
Therefore
Div(X)/Diva(x) = Cl(X)/ClO(X) =?L .
A generalization of this is the following theorem, which was proved by
Severi (for fields of characteristic zero) and Neron (in general).
Theorem D.. For a smooth projective variety X the group
Div(X)jDiya(X) is finitely generated.
It can be shown that for X = II IPn i algebraic equivalence of divisors
is the same thing as equivalence. This example shows that the group
Div(X)/Diva(x) can be more complicated than ?L.
In the case ofa plane cubic curve X the group ClO(X) = Diva(X)/P(X),
where P(X) is the group of principal divisors, is a one-dimensional
Abelian variety. Similarly, for every projective smooth variety there
exists an Abelian variety G whose group of points is isomorphic to
Diva(x)/p(x), and which has the following property. For every algebraic
family of divisors f on X with basis T there exists a regular mapping
cp : T ~ G such that f(t) - f(to) E cp(t), where to is some fixed point of
T [and G is identified with Diva(X)/P(X), hence cp(t) is regarded as a
class of divisors].
The Abelian variety G is uniquely determined by this property.
It is called the Picard variety of X.
The Picard variety of a smooth projective algebraic curve X is also
called its Jacobian variety.
Exercises
1. Let G be an algebraic group, 1p:G x G-G a regular mapping defining a group law,
e e the tangent space to G at the unit element, e~ the tangent space of G x G at the unit
element. Show that e~ = e e E9 e eo and that de 1p : e e E9 e e - e e is given by vector addition.
2. In the notation of Exercise 1, let G be a commutative group and (fJ.: G -+ G be

given by (fJ.(g) = g'. Assuming that the characteristic of the ground field is zero, show
that d.(fJ.(x) is a non-singular linear transformation. Deduce that in a commutative
algebraic group the number of elements of order n is finite and that an n-th root can be
extracted from any element.
§ 4. Differential Forms
1. One-Dimensional Regular Differential Forms. In Ch. II we have intro-
duced the concept of the differential dJ of a function f that is regular
at a point x of an algebraic variety X. By definition dJ is a linear form
on the tangent space ex of x, so that dxf E e!. We now investigate
how this notion depends on the point x.
If the function f is fixed and regular on the whole of X, then dJ in
its dependence on x is an object of a new type we have not met so far:
it associates with every point x E X a vector of the space e! dual to the
tangent space at this point. Later we shall all the time come across ob-
jects of a similar kind. The following explanation may help. In linear
algebra we are concerned with constants, but also with other quantities:
vectors, linear forms, and tensor products. In geometry the analogue to
constants are functions (whose values are constants). The analogues of
vectors, linear forms, etc. are "functions" associating with every point x
of an algebraic (or differentiable) variety X a vector, a linear form, etc.
in the tangent space ex at this point.
We consider the set 4>[X] of all mappings cp that associate with
every point x E X a vector cp(x) of the space e!. Of course, this is far too
large a set, just as the set of all functions on X with values in k is far too
large to be interesting. Similarly to the way in which among all functions
we have selected the regular ones, so we select in the set 4>[X] a part
that is more closely connected with the structure of the variety X. For
this purpose we observe that 4>[X] is an Abelian group if we set
(cp + 1p) (x) = cp(x) + 1p(x). Furthermore, 4>[X] becomes a module over
the ring of all functions on X with values in k if we set (f-cp)(x)= f(x)·cp(x)
for a function f on X and for cp E 4>[X]. In particular, we can regard
4>[X] as a module over the ring k[X] of all regular functions on X.
As we have seen, every function that is regular on X determines a
differential dJ E 4>[X].
Therefore every function f E k[X] determines a function
cp E 4>[X] : cp(x) = dxf, which we denote by df
Defmition. An element cp E 4>[X] is called a regular differential form
on X if every point x E X has a neighbourhood U such that the restric-
tion of cp to U belongs to the submodule of 4>[U] that is generated over
k[U] by the df, f E k[U].
§ 4. Differential Forms 157
Obviously, all regular differential forms on X form a module over

k[X], which we denote by Q[X]. Thus, cP E Q[X] if in a neighbourhood
of every point x E X there is a representation
(1)
where fl' ... , fm' g 1> ••. , gm are regular in a neighbourhood of x.

Taking the differential of a function determines a mapping
d: k[X] ~Q[X]. The properties (1) of Ch. II, § 1.3 now assume the
form
d(f+g) = df+ dg, dU' g) = f· dg +g' df· (2)
From these formulae it is easy to derive an identity, which is true for
any polynomial FE k[Tb ... , T"J and any functionsfl' ... ,fmE k[X]:
m of
d(FUI, ... ,fm))=i~l 07; UI,···,fm)dfi· (3)
For this purpose we have to reduce the proof to the case of a monomial,
using (2), and then prove it by induction on the degree of the monomial,
again using (2). The details of this verification are left to the reader.
Once (3) has been proved for polynomials, it generalizes immediately
to the case of rational functions F. Here we have to keep in mind that
if a rational function F is regular at x, then so are all the functions
of/o7; at this point. For then F = P/Q, where P and Q are poly-
nomials and Q(x) # O. Therefore
i~_ =Q-2(QiP _p_oQ):

07; 07; a7;
from which its regularity follows.
Example 1. X = N. Since at every point x EN the differentials of the
coordinates dxt l , ... , d.Jn form a basis of the space e~, every element
n
cP E cP[An] has a unique representation in the form cP
.
= I I lPidti, where
i=
lPi are functions on N with values in k.
If cP E Q[N], then the decomposition (1) holds in a neighbourhood
of every point. Applying (3) to the g; we obtain the decomposition
cP = :2:: h; dt i, in which the hi are regular at x. Since such a representation
is unique, the lPi must be regular at every point x EN, so that
lPi E keN]. Therefore Q[N] = EB keN] dt;.
" " "
.
Example 2. Let X = 1P 1 and lett be a coordinate on X.
Then X = Ab u A\, with Ab ~ All ~ AI. By the result of Example 1
every elementcpE.Q[p 1 ] can be represented in the form cp=P(t)dt on
A1, qJ = Q(u) du on A\, where ut = 1. From the last relation it follows

that du = - dt/t 2, and in.1A.6 nAt we have
Q(t - 1) Q*(t)
P(t) dt = - --2- dt, that is, P(t) = - 1i+2'
t t
if degQ = n. Here Q*(t) = t n Q(1/t) and Q*(O) #0.
Such a relation among polynomials is possible only when P = Q = O.
Therefore Q[IP1] =0.
Example 3. Let X be given by the equation x~ + xi + x~ = 0 in ]p2
and k be of characteristic different from 3.
We denote by Uij the open set in which Xi # 0, xj # O. Then
X = UOI U U12 U U20 . We set
Xl X2 dy
in UOI :X=-, y=-, qJ=-2'
Xo Xo X
Xz Xo dv
in U12 :u = - , v=-, 1p = 7'
Xl Xl
Xo Xl dt
in UZO:S = - , t = - , X=---y·
Xz Xz S
Evidently qJ E Q[U01 ], 1p E Q[U IZ ], XE Q[Uzo ]. It is easy to verify

that qJ = 1p in Uo1n U1Z , qJ = X in UoIn Uzo . Therefore these formulae
determine a single form ill E Q[X]. This example is interesting in that
Q[X] # 0, whereas X is a projective variety, and there are no non-
constant regular functions on it.
In the general case we can prove a fact that is analogous to, but
weaker than, that in Example 1.
Theorem 1. Every simple point X of an algebraic variety X has an
affine neighbourhood U such that the module Q[U] is free over k[U]. Its
rank is equal to dim xX.
Proof. Let X C AN and suppose that Fl , ... , Fm form a basis of the
ideal of X. Then Fi= 0 on X, and therefore by (3),
N of
I ~dtj=O. (4)
j=1 u~
If X is a simple point and dim xX = n, then the rank of the matrix
((oFJo1j)(x)) is equal to N - n. Suppose, for example, that t l , ... , tn are
local parameters at x. Then it follows from (4) that all the dtj can be
expressed in terms of dtl> ... , dt n with coefficients that are rational
functions and regular at x.
We consider a neighbourhood U of X in which aU these functions
are regular. Then dy!1, ... , dytn form in it a basis of e:
for every point
y E U. Let <p E .Q[U]. By what we have said above, in U there is a

unique representation
n
<p = L lPi dt i , (5)

i= 1
where the lPi are functions on U with values in k. From the repre-
sentation (1) and the formula (3) it follows that <p can be expressed in a
neighbourhood of every point y E U as a linear combination of
dt l , ... , dtN whose coefficients are functions regular in U. As we have
seen, dt l , .•. , dtN can similarly be expressed in terms of dt l , ... , dt n.
n
Therefore <p = L gi dt i, where the gi are regular in a neighbourhood
i= 1
of y. From the uniqueness of the representation (5) it follows that
lPi = gi in a neighbourhood of y, and hence that lPi E k[U]. So we see
n
that .Q[U] = L k[U] dt i •
i=l
Let us assume that among the dt l , ... , dt n there is a relation
n
L gi dt i = 0 and that gIl # 0, say. Then dt l , •.• , dt" are linearly dependent
i= 1
in the open set where gn # 0, but this contradicts the fact that the dyti
are independent in e~ for all y E U. This proves the theorem.
Corollary. If U l , •.. , Un is any system of local parameters at x, then
in some neighbourhood U of x the differentials du l , .•. , dUn generate
the module .Q[U].
Let dtl> ... , dt n be a basis of the free module .Q[U] in a neighbour-
n
hood U of x, which exists according to Theorem 1. Then dUi = L gij dt j,
j= 1
and since the U i are local parameters, Igij(x)1 # O. Therefore ina neighbour-
hood U' in which Igijl # 0 the du\> ... , dUn generate the module .Q[U'].
2. Algebraic Description of the Module of Differentials. We have seen

in Ch. i that the category of affine varieties is equivalent to the category
of rings of a special type. Therefore we can view the whole theory of
affine varieties from a purely algebraic angle, and in particular, we can
try to grasp the algebraic meaning of the module of differential forms.
Consider an affine variety X and denote by A the ring k[X] and by.Q
the module .Q[X]. Taking a differential determines a homomorphism
of k-modules d: A ~.Q.
Proposition 1. The module .0 is generated over A by the elements
df, fEA.
This is an analogue to Theorem 4 of Ch. I, § 3 and is proved in the

same way. If W E Q, then by definition for every point x E X there exists a
representation W = L.fi,xdgi,x, fi,x, gi,x E (!)x' For every function u E (!)x
there exists a representation u=v/w,v,wEA,w(x)#O. Utilizing such a
representation for Ax and gi,x and taking the least common denominator
of all the fractions, we obtain a function Px such that Px(x) # 0,
Since pAx) # 0, there exist functions qx E A such that L PAx = 1, hence

W = L qxri,xdhi,x' This proves Proposition 1.
Proposition 1 suggests the idea of describing the module Q in
terms of its generators df, f EA. Clearly the following relations
hold:
dU + g) = df + dg , df 9 = f dg + gdf
dex = °for ex E k .
(1)
Proposition 2. ~f X is a smooth affine variety, and A = k[X], then

the A-module Q is determined by the relations (1).
Proof We denote by R the module defined over A by generators df
in one-to-one correspondence with the elements of A, and by the rela-
tions (1). There is an obvious homomorphism ~: R --> Q, and Proposi-
tion 1 shows that ~ is an epimorphism.
It remains to show that the kernel of ~ is trivial. Let q> E Rand
~(q» = 0. Observe that the arguments in the proof of Theorem 1 use
only the relations (1). Therefore they are applicable to the module R
°
and show that for every point x E X there exists a function DE A such
that D(x) # and D . q> = Lgidti, gi E A, where now the local parameters
ti are chosen to be elements of A. If ~(q» = 0, then L gidti = in the °
module Q, and from Theorem 1 it follows that all the gi = 0. Thus,
D . q> = 0. So we see for every point x there exists a function DE A, such
that D(x) # 0, D . q> = 0. Arguing as in the proof of Proposition 1 we find
that q> = 0, and the proposition is proved.
Thus, in this case the module Q[XJ can be described purely
algebraically, starting out from the ring k[X]. This suggests the idea of
considering a similar module for every ring A that is an algebra
over a subring Ao. The module R determined by the generators da
and the relations (1) (of course, ex E Ao in the latter), is called the
module of differentials of the ring A over A Q •
If the variety X is not smooth, then this module of differentials R,
which is defined purely algebraically, does not, in general, coincide
with Q[X]. (See Exercise 9). Proposition 1 which is also true for non-
smooth varieties shows that R contains more information on X than the
module Q[X]. However, later we shall be concerned essentially with

smooth varieties, and this difference will not be important for us.
3. Differential Forms of Higher Degrees. The differential forms we
have considered in § 4.1 associate with every point x E X an element of
tbe space e~. Now we consider more general differential forms, which
associate WIth a point x E X a linear skew-symmetric form on the space
ex, that is, an element of the r-th exterior power Are~ of e~.
The definition is entirely analogous to that in § 4.1. We denote by
4>'EX] the set of all correspondences between points x E X and elements
of A'" e~. Thus, if WE 4>r[X], x E X, then w(x) E A'" e~. In particular,
4>°[X] is the ring of arbitrary mappings X -+k; 4>1[X] is the 4>[X]
considered in the previous subsection. Therefore df E 4>1 [X] for
fEk[X].
We recall that the operation of exterior multiplication A is defined
for every vector space L: if cP E A'" L, tp E A'L, then cp A tp E A'"+s L, further-
more, cp A tp is distributive, associative, and tp A cp = ( - 1)" cp A tp. If
e1, .•. , en is a basis of L, then a basis of A'" L consists of all products
eh A ... A ei r , i1 < i2 < ... < ir• Therefore dim A'" L = (~), in particular,
dimAnL= 1, A'" L=O for r>n.
Let us define an operation of exterior multiplication in the sets
4>'EX]: for Wr E 4>'EX], Ws E 4>'[X] we define W = wr A Ws by the equation
w(x) = wr(x) A ws(x) for all x E X. Evidently WE 4>r+s[x]. For r = 1,
s = 0 we arrive at the multiplication of elements of 4>1 [X] = 4> [x] by
functions. Setting s = 0, r arbitrary, we see that a multiplication of
4>'[X] by functions on X is defined. In particular, all the 4>'Ex] are
modules over the ring k[X].
Defmition. An element cp E 4>r [X] is said to be an r-dimensional reg-
ular differential form on X if every point x E X has a neighbourhood
U such that on U the element cplu belongs to the submodule of 4>'[U]
generated over k[U] by the elements df1 A ... A dfr' f1' ... ,fr E k[U].
All the r-dimensional regular differential forms on X form a module
over k[X], which we denote by Q'EX]'
Thus, an element WE Qr[X] can be expressed in a neighbourhood
of an arbitrary point x E X in the form
(1)
where the gi, ... ir , h" "',h r are regular at x.

The operation of exterior multiplication is defined for regular
forms, and obviously for Wr E Q'EX], Ws E Q'[X] we have
Wr A Ws E Qr+s[x]. In particular. every Q'Ex] is a module over k[X].
The differential forms we have considered in the preceding subsection

are, from the point of view of the new definition, one-dimensional.
Theorem 1 has an analogue for forms in QTx] for every r.
Theorem 2. Every simple point of an n-dimensional variety has a
neighbourhood U such that the module Q"[U] isfree over k[U] and of rank
(;) .
Proof In the proof of Theorem 1 we have seen that there exist a
neighbourhood U of a simple point x and It functions U I , ••• , Un> regular
in U, such that dyUI' ... ,dyun form a basis of e;
for every ye U. Hence
it follows that every element cp e 4>'"[U] can be represented in the form
cp = I: '!pi! ... ir du i! /\ ... /\ dUi r
where '!pi! ... ir are functions on U with values in k.
If cp e QTU], then cp can be represented in the form (1) for every
point ye U. Applying Theorem 1 to the forms dU i we see that the functions
'!pi, ... ir are regular at y. Since y is an arbitrary point on U, they are
regular in U. Thus, the forms dUi! /\ ... /\ dUir' i l < i2 < ... < i" generate
the module Q'[U]. It remains to show that these forms are linearly
independent over k[U]. But every dependence
I:gi, ... irduit /\ ... /\duir=O
.
gives at x e U the relation
(2)
Since dxUI' ... , dxu n is a basis of the space e~, we see that the
dxu i, /\ ... /\ dxu ir form a basis in A' e~. Therefore it follows from (2)
that gil "'il(x)=O for all xe U, that is, gi, ... ir=O.
Of particular importance is the module Qn[u], which under the
assumptions of Theorem 2 is of rank 1 over k[U]. Thus, if Q) e Qn[u],
then
(3)
The expression for Q) in this form depends essentially on the choice of
the local parameters Ui ' ... , Un' Let us clarify what this dependence is.
Let Vi' ... , Vn be another It regular functions on X such that
VI - Vi (x), ... , Vn - vn(x) are local parameters at any point x e U. Then
.0 1 [U] = k[U] dV I + ... + k[U] dVn

and, in particular, all the dU i are representable in the form
n
dU i = L hijdvj (i = 1, ... , It). (4)
j=O
§4. Differential Forms 163
Since dxUI, ... , dxu n form a basis of e~ for all x E U, it follows

from (4) that det(hij(x)) ¥= O. By analogy with analysis det(h i ) is called the
Jacobian function of UI, ... , Un with respect to v I, ... , Vn. We denote it by
J (UI' ... , Un) As we have seen, J (U 1 , ••• , Un) E k[U] and for all x E U
VI'···'Vn VI'···'Vn
J(UI' ... 'Un)(X)¥=O. (5)
VI' ••• , Vn
Substitution of (4) in the expression for wand a simple calculation

in the exterior algebra show that
W=g·J (
Ut> ••• , Un) d Ad
VI A ... /\ Vn • (6)
VI' ... , Vn
Thus, although the form WE Qn[u] is given by a function 9 E k[X],
such a specification is possible only when a choice is made of local
parameters and depends essentially on this choice.
We recall that, as a rule, a representation (3) is possible only locally
U
(see Theorems 1 and 2). If X = Ui and if in each Ui such a representation
is possible, it may happen that we cannot associate with W a single
function 9 on the whole of X: the functions gi obtained in the various
Ui need not agree. We have seen an· example of this in §4.1 (Example 3).
4. Rational Differential Forms. Example 2 in § 4.1 shows that on an

algebraic variety X there may be very few regular differential forms
(QI[1PI] = 0), whereas there are plenty of open subsets on which there are
many such forms (Q1[U] = k[U]du). We have come across a similar
phenomenon in connection with the notion of a regular function, and
starting out from just these arguments we have introduced the concept
of a rational function as a function that is regular on some open subset.
We now introduce a similar notion for differential forms.
We consider a smooth irreducible quasiprojective variety X. Let w
be an r-dimensional differential form on X. We recall that it makes sense
to speak of w vanishing at a point XEX:W(X)EAre~, and in parti-
cular, it may be zero.
Lemma. The set of points at which a regular differential form w
vanishes is closed.
Let Y be the set of zeros of the form w. Since closure is a local
property, we may restrict our attention to a sufficiently small neigh-
bourhood U of an arbitrary point x E X. In particular, we may choose U
so that Theorems 1 and 2 hold in it. Then there exists functions
Ul> ... , UnE k[U] such that Q'[U] is a free module with the generators
du;. /\ ... /\ dUir' i l < ... < ir • Therefore W has a unique representation
in the form w = :E gil ... ir du;. /\ ... /\ dU ir and w(x) =0 is equivalent to the
equalities g;. ... ;.(x) = 0, which determine a closed set.
From the lemma it follows, in particular, that if w(x) = 0 for all
points x of an open set U, then w = 0 on the whole of X.
We now introduce a new object, which consists of an open set
U C X and a differential form WE ,g'EU]. For such pairs (w, U) we
define an equivalence relation (w, U) "" (w', U') if w = w' on Un U'. By
the remark made above it is sufficient to require that wand w' agree
on some open set contained in U and U', and the transitivity of this
equivalence relation follows from this. The class defined by it is called a
rational differential form on X. The set of all r-dimensional rational
differential forms on X is denoted by .Q'(X). Clearly ,go (X) = k(X).
Operations on representatives carryover to classes and define a
multiplication: if Wr E ,gr(x), Ws E ,gS(X), then Wr /\ Ws E ,gr+s(x). For s = 0
we see that ,gr(x) is a module over k(X).
If a rational difTerentional form w (which is an equivalence class
of pairs) contains a pair (ii), U), then w is called regular in U. The union
of all open sets in which w is regular is an open set UC!)' the so-called
domain of regularity of w. Evidently w determines a certain regular
form belonging to ,g'EUC!)]. If x E UC!)' then we say that w is regular at x.
Obviously ,gr(x) does not change when X is replaced by an open subset,
in other words, it is a birational invariant.
Let us clarify the structure of the module ,gr(X) over k(X).
Theorem 3. ,gr(x) is a vector space of dimension (~) over k(X).

We consider any open set U C X for which the module ,g'EU] is
free over k[U] (Theorems 1 and 2). Then there exist n functions
u l , ... , Un E k[U] such that the products
(1)
form a basis of ,g'EU] over k[U]. Every form w' E ,gr(X) is regular in
some open set U' C U for which, as before, the forms (1) give a basis in
,g'EU'] over k[U'], Therefore w' is uniquely representable in the form
L
1 S,il < ... <ir~n
gil'''ir du il /\ ... /\ dUi r '
where the gil'" ir are regular in some open set U' C U, that is, rational
on X. But this means that the forms (1) are a basis of ,gr(x) over k(X).
When do n functions U\ • ••• , Un E k(X) have the property that
dUi! /\ ... /\ di;.(1 ~ i l < ... < ir ~ n) is a basis of ,gr(X) over k(X)? We
derive a sufficient condition for this. The condition is also necessary,
but we do not need this.
§4. Differential Forms 165
Theorem 4. If U l , ... , Un is a separable transcendence basis of k(X),

then the forms du;! 1\ ... 1\ dU;r' 1 ~ i l < ... < ir ~ n, ~re a basis of Q"(X)
over k(X).
Since Qr(x) and k(X) are birational invariants, we may take X to
be affine: X C/AN •
Let U l , ... ; Un be a separable transcendence basis of k(X). Then
every element vEk(X) satisfies a relation F(v,u l , ... ,un)=O that is
separable with respect to v.
In particular, the relations Fi(t;, U1, ... , Un) = O(i = 1, ... , N) hold for
all coordinates t; in /AN. It follows from them that
n
Fi.ti dt; + I Fi.Uj dUj= 0 (i = 1, ... , N) on X.

j=l
,From the separability of the polynomials F; with respect to t; it follows
that Fi t. =1= 0 on X. Therefore
.. dt i = f (- F:;uj)
j=l Fi,li
du j • (2)
On some open set U C X all the functions - F:' u/F:' Ii and Ui are regular,
and then (2) shows that at every point Y E U the differentials dyuj
generate e~. Since the number of these differentials is equal to the
dimension of the space, they form a basis of it. Therefore the dUi form
a basis of the module Ql [U] over k[U], the products (1) form a basis
of Q'Eu] over k[U], and a fortiori a basis of Q'(X) over k(X).
Exercises
1. Show that on the affine circle with the equation x 2 + y2 = 1 the rational differential
form dx/y is regular. It is assumed that the characteristic of the ground field is not 2.
2. In the notation of Exercise 1 show that £11 [X] = k[X] ~. Hint: Write an
y
arbitrary form WE = f· ~ and use the fact that ~ = _
£11 [X] in the form W dy .
y y x
3. Show that in Example 3 of § 4.1 dim£11 [X] = 1.
4. Show that £1n[JPn] = O.
5. Show that £11 [lPn] = O.
6. Show that £1'EJPn] = 0 for r > O.
P(t) . .
7. Let W = - - dt, where P and Q are polynomials degP = m, deg Q= n, be a ratIOnal
Q(t)
form on JPI (t is the coordinate on JPI). At what points x E JPI is the form W not
regular?
8. Show that the tangent fibre of the smooth variety X introduced in Ch. II, § 1.4, is
birationally isomorphic to the direct product X x IAn. Hint: For the open set U in
Theorem 1 construct an isomorphism of the tangent fibre to U onto U x R:
(x, ~)->x x «(dxUI) (~), ... , (dxu n) (~)), ~ E ex'
9. Compute the module R constructed in the proof of Proposition 1 in § 4.2 for the
curve y2=X 3 , and show that ~(3ydx-2xdy)=O. Hint: Use the fact that
k[X] = k[x] + k[x]y.
10. Let K be an extension of k. A derivation of Kover k is a k-linear mapping

D : K -> K satisfying the condition D(xy) = D(x) Y + xD(y), x, Y E K. Show that if u E K
and if D is a derivation, then so is the mapping D , (x)=uD(x), so that all the derivations
of Kover k form a vector space over K, which is denoted by Dk(K).
11. Let D be a derivation of the field K = k(X) over k, co E Q1 (X), co = r. Ii dg i. Show
that the function (D, co) = r. J.D(gi) does not depend on the representation of co in the
form r. J.dg i. Show also that this scalar product establishes an isomorphism
Dk(K) ~ (Q1 (X))* = Hom k(x)(Q' (X), k(X)).
§ 5. Examples and Applications of Differential Forms

1. Behaviour under Mappings. We begin by investigating the behaviour
of differential forms under regular mappings. If cp: X --> Y is such a
mapping, XEX, then dxcp is a mapping ex,x-->e,p(x),y, and the adjoint.
transformation (dxcp)* maps e*(x), y into e~,x' Hence for WE cJ>[Y] we
have Cp*(W)EcJ>[X], where cp*(w) (X)= (dxcp)* (w (cp(x))).
From the definition it follows easily that the mapping (dxcp)* is
compatible with taking the differential, that is, cp* (drp(x)f) = dx(cp*(f))
for f E keY]. Hence it follows that if WE Ql [y], then cp*(w) E Ql [X],
and cp* determines a homomorphism cp*: Ql [Y] -->Q 1 [X], which is
compatible with taking the differential for f E k[ Y].
Finally, from linear algebra it is standard knowledge that a linear
transformation of linear spaces cp: L--> M determines a linear transform-
ation Arcp:ArL-->ArM. Applying this to the mapping (dxcp)* we obtain a
mapping Ar(dxcp)*: AYe:(xJ,y-->Are~.x and mappings cJ>'[Y]-->cJ>Y[X]
and Qr[y] -->QY[X]. We denote the latter again by cp*.
From all we have said above it follows that the effective computation
of the action of the operator cp* on a differential form is very simple: if
W = L gi, .. irdui, /\ ... /\ dU ir
then
cp*(W) = L cp*(gi, ... J d(cp*(u i ,)) /\ ••. /\ d(CP*(UiJ)· (1)
Now let X be irreducible, cp: X --> Y a rational mapping, and cp(X)
dense in Y. Since cp is a regular mapping of an open set U C X into Y
and every open set V C Y intersects cp(U), the preceding arguments
determine a mapping cp*: QY(Y)-->Q'·(X). This mapping is again given
by formula (1).
We know that for r = 0, that is, for functions, the mapping cp* is an
embedding. For differential forms this is not always the case. For
example, let X=Y=lP 1 , k(X) = k(t), k(Y)=k(u), let k be of finite
characteristic p, and let cp be given by the formula u = tP • Then
cp*(j(u))=f(tP) and cp*(df)=d(j(tP))=O, (jEk(u)), so that cp*(Ql(y))=O.
The situation is clarified by the following result.
Theorem 1. If the field k(X) has a separable transcendence basis

over k(Y), then the mapping q>*: Qr(Y)-+Qr(x) is an embedding.
Here we identify the field keY) with the subfield q>*k(Y) of k(X).
Suppose that k(X)/k(Y) has a separable transcendence basis
v l' ... , Vs' This means that VI"'" Vs are algebraically indep.endent
over k(Y), and that k(X) is a finite separable extension of k(Y) (Vl' ... , vs).
The field keY) has a separable transcendence basis over k (see Note 1
to Theorem 6 of Ch. I, § 3), which we denote by U 1 , ••. , U r • Then
Ul' ... , U" v 1, ... , Vs is a separable transcendence basis of k(X) over k.
. When we write an arbitrary differential form WE Qr(Y} as
W = :E g;. ... i, dU i1 /\ ... /\ du i, (2)
and apply (1) to it, we obtain an expression for q>*(w) in terms of
products dq>*(u i.) /\ ... /\ dq>*(uiJ, which form part of a basis of Qr(x)
over k(X), because q>*(uJ is part of the separable transcendence basis
U1, ... ,Ur, V1"",V s (Theorem 4 of§4). Therefore q>*(W) =0 only if all
the q>*(gil ... iJ=O, but this is possible only when g;' ... i,=O, that is, w=O.
All the preceding results were more or less obvious. Now we come
to an unexpected fact.
Theorem 2. If X and Yare smooth varieties, Y projective, and
q> : X -+ Y a rational mapping such that q>(X) is dense in Y, then
q>*Qr[y] C Q'[X].
In other words, q>* carries regular differential forms into regular
ones. Since q> is only rational, this seems altogether improbable even
for functions, that is, for r = O. In this case the situation is saved by the
fact that owing to Y being projective, regular functions on Yare
constants, and the theorem is trivial.
In the general case the theorem is less obvious. We utilize the fact
that by Theorem 3 of Ch. II, § 3, the mapping q> is regular on X - Z,
where Z is closed in X and codimxZ ~ 2. If WE QTY], then q>*(w) is
regular on X - Z. Let us show that this implies its regularity on the
whole of X. For this purpose we write q>*(w) in an open set U in the
standard form (2) [with W replaced by q>*(w)], where now U 1 , ••• , Un are
regular functions on U such that du;,/\ ... /\ dUi, is a basis of QTUJ
over k[U]. Then the regularity of the forms q>*(w) on X - Z implies
the regularity of all the functions g;I"';' in U - (Z n U). But
codimu(ZnU)~2, and this means that the set of points where the
g;I'" i, are not regular is of co dimension ~ 2. On the other hand, this
set is a divisor (g;. ... ;)oo' This is possible only when (g;. ... ;)oo =0 and
hence the function gil"';' is regular.
Corollary. If two smooth projective varieties X and Yare birationally
isomorphic, then the vector spaces Q'[X] and Q'[Y] over k are isomorphic.
The significance of Theorem 2 and its corollary is enhanced by the

fact that for a projective variety X the space Q'[X] is finite-dimensional
over k. This result is a consequence of a general theorem on coherent
sheaves, which will be proved in Ch. VI. For the case of curves we
prove it in Subsection 3. We set hr =dimQTX]. The corollary to Theorem
2 indicates that the numbers hr(r = 0, 1, ... , n) are birational invariants
of a smooth projective variety X.
2. Invariant Differential Forms on a Group. Let X be an algebraic

variety, 0) a differential form of it, and g an automorphism of X. The
form 0) is said to be invariant under g if
g*(O)) = 0) .
In particular, let G be an algebraic group. From the definition

given in § 3.2 it follows at once that for every element g E G the mapping
tg(x) = g. x
is regular and is an automorphism of G qua algebraic variety.
A differential form on G is called invariant if it is invariant under all
the transformations t g •
An invariant differential form is regular. For if a form 0) is regular
at a point xoEG, then t;O) is regular at g-lx o. But t;O)=O), hence 0) is
regular at all points gxo, g E G, and these are all the points of G.
We show how to find all the invariant differential forms on an
algebraic group. For this purpose we observe that a mapping
f: X ~ Y determines a mapping f* not only on differential forms, but
also on the vector spaces rpr:
f* : rpr[Y] ~ rpr[x] .
t;
In particular, the are automorphisms of the vector spaces rpr[G]. We
begin by determining the set of elements ep E rpr[G] that are invariant
under all t;, g E G. This set contains, in particular, the invariant
differential forms.
The condition
t;(ep) = ep
means that for every point x E G
ep(x) = (Ar dt;)(ep(gx)). (1)
In particular, for g = x - 1,
(Ar dt;- d (ep(e)) = ep(x) . (2)
This formula shows that q> is uniquely determined by' the element q>(e)
of the finite-dimensional vector space A r e:.
Conversely, by specifying
an arbitrary 11 E Are:, we can construct by (2) an element q> E IP'CG]:
q>(x) = (Ar dt~-.) (11).
t:.
A simple substitution shows that it also satisfies (1), in other words,
is invariant under
t:
Thus, the subspace of elements q> E IP'CG] that
are invariant under the automorphisms is isomorphic to ArIP:, and
the isomorphism is given by the correspondence
q>-+q>(e) •
Let us now show that all the elements q> we have constructed are
regular differential forms, that is, contained in Q'C G]. Owing to the
invariance, regularity of a form q> need only be verified at an arbitrary
single point, for example, at the unit point e. Furthermore, we may
restrict ourselves to the case r = 1. For if
11 = L !Xii 1\ ... 1\ !Xir' !Xj E A! e: ,
and if the forms q>j corresponding to the !Xj by (2) are regular, then the
form q> = L q>il 1\ ... 1\ q>i r is regular and corresponds to 11.
We take an affine neighbourhood V of e such that the module
Ql [V] is free, and let du!, ... , dUn be a basis of it. There exists an affine
neighbourhood U of e such that J,l(U x U) C V, where J,l is the mapping
that defines multiplication in G. Like every function in k[U x UJ, J,l*(u l )
can be written in the form
J,l*(UI)(g!, g2) = L vlj(g d WIj(g2)' Vlj' W/j E k[U] ,
(g! , 9 2) E V X U CG x G .
By definition th = J,lSh, where Sh is the embedding G -+ G x G, Sh(g) = (h, g).
Therefore
(t; dUI) (g) = L vlj(g) dew, j •
When we express dWI. in terms of the dUk, we obtain the relations

J
t;dul= LCml(g)dum, CmlE k[U] , (3)

where
cmM) = L Vlj(g) ----:;-L
j
OWl'
vU
(e) . (4)
m
Now we write the invariant form q> as q> = LlPmdu". and consider
t;
the relation q> = q> at e. Substituting the expressions (3) and equating
coefficients of dUm' we obtain
(5)
Since (cml(e») is the unit matrix, we have det(cml) (e) =1= 0, and from the
system of Eqs. (5) it follows that "PmE (!Je.
Let us state the result we have proved:
Proposition. The mapping w--+w(e) establishes an isomorphism
between the space of r-dimensional invariant regular differential forms
on G and the space Ar e:.
3. The Canonical Class. Now we make a special analysis of n-dimen-
sional rational differential forms on an n-dimensional smooth variety X.
In some neighbourhood of a point x E X such a form can be represented
as w = g dUl 1\ ... 1\ dUn. We cover the whole of X with affine sets Ui
such that in each of them this representation w = g(i) duY) 1\ ... 1\ du~)
holds. In the intersection Ui n Uj we find, according to (6) of § 4.3, that
ti) = (i») ( uIi)

1 ,
(i»)
••• , Un
g g (j) (j).
U 1 , •.. , Un
Since the Jacobian) is regular and non-zero in UinUj [see (5) in §4.3],
the system of functions g(i) in Ui is compatible in the sense of § 1.2 and
therefore determines a divisor on X. This is called the divisor of the
form wand is denoted by (w).
The following properties of a divisor of an n-dimensional differential
form on an n-dimensional variety follow easily from the definition:
a) (f. w)=(f) +(w) if fE k(X);
b) (w) ~ 0 if and only if WE nn[x].
According to Theorem 3 of § 4 (for r = n) the space nn(x) is one-
dimensional over k( X). Therefore, if Wi E nn(x), Wi =1= 0, then every
form WE nn(x) can be represented as w = f Wi. Hence property a) shows
that the divisors of all the forms WE nn(x) are equivalent to each other
and form a single divisor class on x.
This divisor class is called the canonical class of X and is denoted
by K or Kx.
Let Wi be a fixed form in nn(x) in terms of which every form can
be expressed as w = f Wi. Property b) shows that w is regular on X if
and only if (f)+(Wl)~O. In order words, nn[X}~.;t'(Wl»)' where we
make use of the concept of the space associated with a divisor that was
introduced in § 1.5.
Thus, hn = dimknn[X] = 1((wl») = I(K). So we see that the invariant
h introduced in § 5.1 coincides with the dimension of the canonical class.
n
Example. Let us assume that X is the variety of an algebraic group.

In § 5.2 we have shown that the space of r-dimensional invariant
differential forms on X is isomorphic to Are:, where ee is the tangent
space to X at the unit point e. In particular, the space of n-dimensional
invariant differential forms is one-dimensional because A ne: ~ k. If W

is a non-zero invariant form, then WE Qn[x], that is, (w);;;, O. But if
w(x) = 0 for some point x E X, then by the invariance also w(y) = 0 for
every point Y E X. Therefore w(x) ¥= 0 for all x E X, that is, w is regular
and does not vanish on X. This means that (w) = 0 or that Kx = O.
In § 2 we have shown that the number ltD) is finite for any divisor D
on a smooth projective algebraic curve. Hence it follows, in particular,
that the number hI = dimkQI [X] is finite for any smooth projective
algebraic curve X. This number is called the genus of the curve and is
denoted by g(X) or g; here hI =g if dimX = 1.
When dimX = 1, we know that all the divisors of one class have one
and the same degree, so that we can speak of the degree deg C of
a class C. In particular, the degree deg Kx of the canonical class is a
birational invariant of the curve X.
The invariants we have introduced: the genus g(X) and degKx are
not independent. It can be shown that they are linked by the relation
deg K x = 2g(X) - 2. (See § 5.6.) In particular, if a smooth projective
curve X is an algebraic group, then K x = 0, as we have just seen.
Therefore gx = 1, that is, among all projective curves only on those of
genus 1 can the law of an algebraic group be defined. We shall see in
§ 5.6 that curves of genus 1 are precisely the smooth cubic curves.
4. Hypersurfaces. Next we compute the canonical class and determine

hn for the case when X is a smooth hypersurface in lPN, n = dim X = N - 1.
Let X be given by an equation F(xo: ... :x N) = 0, degF = degX = m. We
consider an affine open set U in which Xo ¥= O. In it X is given by an
equation G(Yt, "',YN)= 0, G(YI' ""YN) = F(I, YI, ""YN)' where Yi=XJXO'
In the open subset Ui C U in which G;, ¥= 0 local parameters are
A
YI""'Yi"",YN, and the form dYI/\···/\dy;!\···/\dYN is a basis of

Qn [Ui] over k [U;]. However, it is convenient to take as a basis the form
1 A
Wi = -G' dYI/\ ... /\ dYi /\ ... /\ dYN
y,
(which is possible, because G~, ¥= 0 in U;). The fact is that the forms
WI' ... , W N are very simply connected with each other: multiplying the
relation
N
L
i= I
G;,dYi=O
A A
by dYI/\'" /\ dYi /\ ... /\ dYj /\ ... /\ dYN we see that
Wj = (- l)i+ j Wi . (1)
Since X is smooth, U = UUi' and it follows from (1) that all the forms
Wj are regular in the whole of U and that the divisor of these forms in U
is equal to 0.
It remains to investigate the points that do not belong to U. Let us
consider, for example, an open subset V in which Xl =I- 0. Coordinates in
this affine variety are ZI, ... ,ZN:Zl=1/YI' Zi=yJYI (i=2, ... ,N).
Evidently
Zi
Yi=Z- (i= 2, ... ,N). (2)
I
Therefore
(i=2, ... ,N).
We substitute these expression in WN. Using the fact that dz I /\ dz I = 0,

we obtain
The equation of X in V is of the form
Zz, ... , -ZN) .

where H=ziG (-1, -
Zl Zl Zl
From the relation
it follows that
(3)
All the arguments referring to U also apply to V and show that
1
,an [V] = k[V]H' dZ l /\ ••. /\ dZ N _ 1 • (4)
ZN
Therefore in V we have (w N ) = - (N - m + 1)· (Zl). Evidently (Zl) in V is

a divisor of the form Xo on X, as it was defined in § 1.2. Ultimately we
find that the relation (w N ) = (m - N - 1) . (xo) = (m - n - 2). (xo) holds
on X. Thus, Kx is the divisor class containing (m - n - 2)L, where L is a
section of X by a hyperplane.
Now let us find ,an[x]. We know that ,an[U]=k[U]WN. Let
W=P(YI, .. ·'YN)WN,PEk[YI' ···'YN].
Substituting (2) and using (3) we find that in V
P(Zl,.··,ZN) 1 d did P
W= - l+N m+l -H' Zl/\"'/\ ZN-l, = eg ,
Zl ZN
From (4) it now follows at once that (w);;;.O in V if and only if

p/zi+ N - m +1 Ek[V],
that is, when I+N-m+1,,;;O or l,,;;m-N-1.
Thus, WE Qn[x] if and only if W = p. WN'
degP,,;;m-N -1 =m-n-2. (5)
Hence it is easy to calculate the dimension of Qn [X]. Namely, two
distinct polynomials P,QEk[Yl' .. ',YN] satisfying the condition (5)
determine distinct elements of the ring k [X], in other words, P - Q== O( G),
and this contradicts (5). Thus, the dimension of Qn [X] is the same as that
of the space of the polynomials P satisfying (5). This dimension is
(m-1) ... (m-N)/N!= ( m-1)
N . Thus,
hn(x) = (mn+1
- 1) . (6)
Here is the simplest case of this formula: for N = 2, n = 1
g(X) = (m -=-1)(m - 2)
2
is the formula for the genus of a smooth plane curve of degree m.
From (6) we can draw at once an important conclusion. Interpreting
(: ~~) as the number of combinations we see that for m> m' > n + 1
(:~~) > (m~~~).

Therefore (6) shows that hypersurfaces of distinct degrees m, m' > n + 1
are birationally non-isomorphic. So we see that there exist infmitely
many birationally non-isomorphic algebraic varieties of a given dimen-
sion.
In particular, for N = 2, m = 3 we obtain g(X) = 1, and since g(1Pl) = 0,
we see that a smooth projective plane curve of degree 3 is non-rational.
From (6) it follows that hn(X) = 0 if m ,,;; N. In particular, hn(JPn) = O.
For n = 1 we have verified this directly in § 5.2.
Let us consider in more detail the case m ~ N. If N = 2, this means

that m = 1 or 2. For m = 1 we have X =]pl and we know already that
hl(1PI)=O. For m=2 we are concerned with a smooth conic, which is
isomorphic to ]p I, so that in this case the equality hi (X) = 0 does not tell
us anything new.
Let N = 3. For m = 1 we are concerned with ]p2, and the equality
h2 = 0 is already known to us. For m = 2, X is a quadric, which is
birationally isomorphic to ]p2, so that the equality h2 (X) = 0 is a conse-
quence of h2(1P2) = 0 and Theorem 2. For m= 3, X is a cubic surface. If
on such a surface there are two skew lines, then it is birationally iso-
morphic to ]p2 (see Example 2 of Ch. I, § 3.3). It can be shown that any
smooth cubic surface contains two skew lines*, so that again the equality
h2(X) = 0 is a consequence of Theorem 2 and the fact that h2(1P2) = o.
These examples lead to interesting questions on smo'oth hyper-
surfaces of small degree: Xc lPN, m = deg X ~ N. We see that for N = 2
or 3 the hypersurface X is birationally isomorphic to the projective space
JPN-I, which gives an "explanation" of the equality h"(X)=O, n=N-l.
For N = 4 we come across a new phenomenon. When m = 3, for
example, even for the hypersurface
~+~+~+~+~=O m
the question whether it is birationally isomorphic to ]p3 is very delicate.
However, it can be shown that there exists a rational mapping qJ:]p3 _ X
such that qJ(]p3) is dense in X and k(]p3) separable over k(X) (see
Exercise 18). Together with the equality h3(]p3) = 0 and Theorem 2 this
gives h3 (X)=O. In this context we introduce the following terminology:
a variety X is called rational if it is birationally isomorphic to lPn,
n = dimX, and unirational if there exists a rational mapping qJ:]P" - X
such that qJ(P") is dense in X and k(1P")/k(X) is separable. From Theorem 2
and Exercise 6 in § 4 it follows that for a unirational variety X all the
hi(X)=O.
Typical for a number of difficulties that occur in algebraic geometry
is the question whether the concepts of a rational and a unirational variety
are one and the same. This is the so-called Liiroth problem. Clearly, it
can be restated as a question in the theory of fields: let K be a subfield
ofthe field ofrational functions k(TI' ... , TJ such that k(TI' ... , T")/K is
finite and separable; is K isomorphic to a field of rational functions?
For n = 1 the answer is in the affirmative, even without the assumption
that k is algebraically closed and that k(T)/K is separable.
For n = 2 without these restrictions the answer is in the negative,
with them it is in the affirmative, but the proof is very subtle. For fields
of characteristic zero there is an account, for example, in [3], Ch. III.
* See cy. Manin, Y. Cubic Forms, North-Holland, 1974, p. 118.
For n;;:. 3 the answer is in the negative even when k is the field of
complex numbers. The simplest example of a unirational but not rational
variety is a 3-dimensional smooth hypersurface of degree 3 in 1P4, in
particular, the hypersurface (7). (See Ex. 18 and footnote to p. 425.)
5. Hyperelliptic Curves. As a second example we consider one type

of curves. We denote by Y an affme plane curve with the equation
y2 = F(x), where F(x) is a polynomial without multiple roots of odd
degree n = 2m + 1 (in Ch. I, § 1, it was proved that the case of even degree
reduces to that of odd degree). We assume that the characteristic of k is
not 2. A smooth projective model X of the curve Y is called a hyperelliptic
curve. We compute the canonical class and the genus of X.
The rational mapping (x, y)-+x of the curve Y in /A1 determines
a regular mapping j : X -+ ]pl. Clearly deg j = 2, so that by Theorem 1 of
§ 2 for oc E]p1 either j -1 (oc) consists of two points z' and z" in each of
which vz-(u) = vz,,(u) = 1 for a local parameter u at oc, or else j-l(OC)=Z
and vz(u) = 2.
The affine curve Y is easily seen to be smooth. If Y is its projective
closure, then X is a normalization of Yand we have the mapping cp :X -+ Y,
which is an isomorphism between Y and cp-1(y). Hence it follows that
if a point ~ E /A1 has the coordinate oc and F(oc),= 0, then j -1(~) = (z', z"),
and if F(oc) = 0, then j -1(~) = z.
Let us consider the point at infmity OC oo E ]p 1. If the coordinate on /A1
is denoted by x, then u= x- 1 is a local parameter at OCoo- Ifj-l(oc oo ) were
to consist of two points z' and z", then at z', say, the function u would be
a local parameter. Hence it would follow that vz·(u) = 1, vz-(F(x») = -no
But since n is odd, this contradicts the fact that vz-(F(x») = 2v z.(y). Thus,
j-1(oc oo ) consists of a single point, which we denote by Zoo, and vz"'(x)
= -2, vz..,(y) = -no It follows that X=cp-1(y)UZoo-
Let us now turn to differential forms on X. Consider, for example,
the form OJ = dx/y. If y(~),= 0 at a point ~ E Y, then x is a local parameter
and v~(OJ)=O. But if y(~)=O, then y is a local parameter and v~(x)=2,
from which it follows again that v~(OJ)=O. Thus, (OJ)=l·zoo, and it
remains for us to determine l. For this purpose we recall that if t is a
local parameter at Zoo' then x = t- 2 u, y= t-nv, where u and v are units in
(9z",' Therefore OJ=rn- 3 wdt, with w, w- 1 E(9z"" hence (OJ) = (n-3)zoo.
Now let us find Q1 [X]. As we have seen, OJ forms a basis of the
module Ql [Y] : Q1 [Y] = k [Y] OJ, so that every form in Ql [X] can be
written as UOJ, where u E k[Y], hence can be represented as P(x) + Q(x)y,
P, QEk[X].
It remains to clarify which of these forms are regular at zoo' This is so
if and only if
vz",(u);;:, -(n-3). (1 )
Let us find such u E k [Y]. Since vzJx) = - 2, vz"'(P(x)) is always even,

and since vzJy) = - n, vz..,(Q(x)y) is odd. Therefore
vzJu) = vz"'(P(x) + Q(x)y),;;;; min (vz..,(P(x)), vz..,(Q(x)y)) ,
hence, if Q # 0, then vzJu),;;;; - n. Thus, u = P(x), and (1) shows that

2degP,;;;;n-3.
We have found that Ql [X] consists of forms P(x) dx/y, where the
degree of the polynomial P(x) does not exceed (n - 3)/2. Hence g = hl
=dimQ 1 [X]=(n-1)/2.
It is intersting to compare the results of § 5.4 and § 5.5 for N = 2. In
the second case we have seen that there exist algebraic curves of any
preassigned genus and in the first that the genus of a plane smooth curve is
of the form (n - 1) (n - 2)/2, in other words, is by no means an arbitrary
integer. Thus, not every smooth projective curve is isomorphic to a plane
smooth curve. For example, this is not true for hyperelliptic curves with
n=9.
6. The Riemann-Roch Theorem for Curves. One of the central results
of the theory of algebraic curves is the Riemann-Roch theorem. It is
expressed in the equation
I(D) - I(K - D) = degD - g + 1 , (1)
where D is an arbitrary divisor on a smooth projective curve, Kits
canonical class, and g its genus.
The proof of this theorem goes deep into the details of algebraic
curves and will therefore not be given here. However, we can indicate
some of its consequences which make its value for theory of curves quite
manifest.
Corollary 1. Setting D = K we find that, since I(K - K) = 1(0) = 1, and
I(K) = g, we have degK = 2g - 2.
Of this equation we have talked in § 5.3.
Corollary 2. If degD > 2g - 2, then I(K - D) =0.
For otherwise there would exist a divisor D' such that K - D '" D' > 0,
but this is impossible because degD' < 0. Thus, the Riemann-Roch
theorem shows that I(D) = degD - g + 1 for degD > 2g - 2.
Corollary 3. If g = ° and D = x is a point on X, then by (1), 1(D) > 2.
This means that the space 5l'(D) contains, apart from constants, also
a nonconstant function f For such a function (ft) = x, that is, if we
in terpret f as a mapping f: X ~ IP 1, then deg f = 1 by Theorem 1 of § 2.
°
Hence it follows that X::= IP l, that is, the equality g = is not only
necessary, but also sufficient for a curve X to be rational.
Corollary 4. We consider a basis fo, ... ,fn of the space 2(D), D -;;;. 0,
and the corresponding rational mapping cp = (fo: ... :fn), X ~ lPn. Let us
clarify when cp is an embedding. We show that this is so under the following
conditions:
I(D-x)=I(D)-1,
(2)
I(D - x - y) = I(D) - 2 ,
for arbitrary points x, y E X.

From Corollary 2 it follows that the Eq. (2) are true if degD -;;;. 2g + 1,
so that in this case cp is an embedding.
We note first of all, that the first conditions in (2) guarantees that
- D = g.c.d. (fJ For by definition g.c.d. (fi) -;;;. - D. Now if we did not
have equality, then there would exist a point x such that (1;) -;;;. - D + x,
that is, 2(D) C 2(D - x), I(D).;;; I(D - x), which contradicts (2). Thus,
according to the remark at the end of § 1.4, the divisors D). = (LA-ih) + D
are inverse images of hyperplanes under the mapping cpo
To prove that cp is an isomorphism we use the lemma in Ch. II, § 5.5,
whose conditions we can verify by means of the remark made above.
If cp(x) = cp(y), then every hyperplane E passing through the point cp(x)
also passes through cp(y). This means that if D). - x -;;;. 0, then
D). - x - y -;;;. 0, that is, I(D - x) .;;; I(D - x - y), which contradicts the
second condition in (2).
Let us show that the tangent spaces are mapped isomorphically.
This is equivalent to the fact that
is an epimorphism. If this is not so, then cp*(m<p(X» em;, because in our

case dimmJm; = 1. In other words, for any function u E m<p(x) we have
vAcp*(u») -;;;. 2. Applied to linear functions this shows that if D;. - x -;;;. 0,
then D). - 2x -;;;. 0. Again we find that I(D- x).;;; I(D - 2x), which contra-
dicts the second condition in (2) and completes the proof.
Obviously, when a different basis is chosen in 2(D), the mapping
cp is multiplied by a projective transformation of the space lPn. On the
other hand, replacing D by another divisor D + (f) leads to an iso-
morphism u~uf of the space 2(D) and therefore does not change cpo
Thus, it makes sense to talk of the mapping cp corresponding to a class of
divisors.
For example, let X be a curve of genus 1, Xo E X. The conditions of
Corollary 4 hold for the divisor 3x o. Therefore the mapping cp correspond-
ing to this divisor maps X isomorphically onto a curve X' C lP 2 [because
1(3x o) = 3 by (1)]. As we have seen, 3x o is the inverse image of a section
of X' by a line, and since deg3xo = 3, also degX' = 3. Thus, every curve
of genus 1 is isomorphic to a plane cubic curve.
Of the greatest interest are mappings CfJ that correspond to classes

intrinsically connected with the curve X. Such are, for example, the
multiples nK of the canonical class. The Eq. (1) shows that degnK
.~ 2g + 1 for n ~ 2 if g > 2, and n ~ 3 if g = 2. Thus, for g> 1 the class 3K
always satisfies the conditions of Corollary 4. The corresponding mapping
CfJ3K maps the curve X into ]pm, where m = 1(3K) - 1 = 5g - 6 (by
Corollary 2). Here two curves X and X' are isomorphic if and only if their
images CfJ3K(X) and CfJ3K(X) are obtained from each other by projective
transformations of the space. In this way the problem of a birational
classification reduces to a projective classification.
The mapping CfJ corresponding to the canonical class is not always
an embedding. However, all the cases when this is not so can be
enumerated (see Exercises 16 and 17).
As a simple application of these arguments we consider plane curves
of degree 4. According to § 5.4 their canonical class coincides with the
class of the intersection with a line in ]p2. Therefore the mapping CfJK
corresponding to the canonical class coincides with their natural embed-
ding in a plane. From what we have said above it follows that two such
curves are isomorphic if and only if they are projectively equivalent. This
leads us to a very important conclusion The set of plane curves of degree 4
can be identified with the space ]p14 (Ch. I, § 4.4). On the other hand, the
group of all projective transformations of a plane is of dimension 8
(matrices of order 3 to within a constant factor). Using the theorem on
the dimension of fibres it is easy to deduce that in ]p14 there exists an
open set U and a mapping f : U -+ M onto some variety M such that two
points U 1 and U2 in U correspond to projectively equivalent curves only
if they lie in one fibre of the mapping f. Therefore the dimension of the
fibre is 8, and dimM = 14 - 8 = 6.
Thus, it is by no means true that any two curves of degree 4 are
isomorphic: they must also correspond to one and the same point on the
6-dimensional variety M. This shows that the genus is not a complete
system of birational invariants of curves. Apart from the integral-valued
invariant, the genus, curves also have "continuous" invariants, the so-
called moduli. It can be shown that all curves of given genus g> 1 form
(in a sense we do not make precise here) a single continuous variety of
dimension 3g - 3. In the case of curves of degree 4 we have g = 3 and
3g-3=6=dimM. A similar also holds for curves of genus 1 (see
Exercises 12 and 13). Only for g = 0 are all the curves of this genus
isomorphic.
7. Projective Immersions of Surfaces. Here we give an account of

how the facts proved in the preceding subsection for algebraic curves can
be generalized to surfaces. No proofs are given. The reader can fmd them
in the book [3]. Furthermore, we restrict ourselves to the case of a field

oJ characteristic O. -
An analogue to curves of genus greater than 1 are surfaces for which
a mUltiple of the canonical class determines a birational isomorphism.
They are called surfaces of general type, and for them a birational
classification reduces in a certain sense to projective classification. The
main result on surfaces of general type consists in the fact that for them
already the five-fold canonical class 5 K determines a regular mapping
and a birational isomorphism.
It remains to list the surfaces that are not of general type. They play
the role of curves of genus 0 and 1 and are given by similar constructions.
Analogues to rational curves are, firstly, rational surfaces, that is,
surfaces birationally isomorphic to IP2, and, secondly, ruled surfaces.
These are surfaces X that can be mapped onto a curve C in such a way
that all fibres of this mapping are isomorphic to a projective line IPl.
Thus, they are an algebraic family of lines.
Analogues to curves of genus 1 are three types of surfaces.
The first type are two-dimensional Abelian varieties. Surfaces of the
second type (which are called K 3 surfaces) have the property, in common
with Abelian varieties, that their canonical class is O. However, in contrast
to Abelian varieties there are no regular one-dimensional differential
forms on them (according to the results in § 5.2 on Abelian varieties there
exist invariant, hence regular, one-dimensional differential forms). The
third type are elliptic surfaces, that is, families of elliptic curves. These
surfaces possess a mapping f :X ~ C onto a curve C such that for all
Y E C for which f -1 (y) is a smooth curve (and such are all the y apart
from finitely many) this curve has genus 1.
The main theorem states that all surfaces not of general type to within
a birational isomorphism are exhausted by the five listed: rational, ruled,
Abelian, K3 and elliptic.
To throw more light on these classes of surfaces it is convenient to
classify them by an invariant x, the maximal dimension of the image of a
surface X under the rational mappings given by the divisor classes nK,
n = 1,2, .... If l(nK) = 0 for all 11, then they are no such mappings and
we set x = - 1. Here is the result of the classification. Surfaces of general
type are those for which x = 2. Surfaces with x = 1 are all elliptic. More
accurately, they are the elliptic surfaces for which nK =1= 0 for n =1= O. The
order of the canonical class of an elliptic surface X in the group CI(X) is
infinite or is a divisor of 12. Surfaces with x = 0 are characterized by the
condition 12K = O. Thus, they are elliptic surfaces for which 12K = 0,
surfaces of type K 3, and two-dimensional Abelian varieties. Surfaces
with x = - 1 are rational or ruled.
F or each of these types there is a characterization in terms of in variants

similar to the way in which g = 0 characterizes rational curves. We quote
the characterization only for the first two types. For this purpose we
use the result of Exercise 7, according to which the numbers l(mK) for
m;;;. 0 are birational invariants of smooth projective varieties. They are
called multiple genera and are denoted by Pm. In particular P1 = hn
= dimQn[X] for n= dimX.
Criterion for Rationality. A surface X is rational if and only if Q1 [X]
= 0 and P 1 = P2 = O.
A solution of Ltiroth's problem for surfaces follows easily from this
criterion.
Criterion for Being Ruled. A surface X is ruled if and only if P3
=P4 =O.
Generalizations of the result reported in this subsection to varieties
of dimension > 2 are not known. *
Exercises
1. Show that df = 0 for an element f E k(X) if and only iff E k (when k is of charac-
teristic 0) or f = gP (when k is of characteristic p > 0). Hint: Use Theorem 1 and the following
lemma: if KjL is a finite separable extension of characteristic p, x E K, and its minimal
polynomial is of the form :r.af Xi, ai E L, then x = yP, Y E K.
2. Let X and Y be smooth projective curves, <p: X -+ Y a regular mapping such that
<p(X) = Y, let x E X, Y E Y, <p(x) = y, and let t be a local parameter at y. Show that the number
ex = Vx (<p* dt) does not depend on the choice ofthe local parameter t and that ex> 0 if and
only if x is a ramification point of <po The number ex is called the ramification multiplicity of x.
3. In the notation of Exercise 2, let <p*(y) = :r.li Xi' where y is the divisor consisting of
the single point y. Suppose that the characteristic of k is either 0 or p> Ii' Show that ex,
=[i-1.
4. In the notation of Exercises 2 and 3, let Y = lP I. Show that g(X) = t L ex - degrp + 1.
XEX
Generalize this relation to the case of an arbitrary curve Y.
5. Suppose that <p: X -+ Y satisfies the conditions of Exercise 2. Show that a differential
WE QI(Y) is regular when the differential <p*w E QI(X) is regular.
6. Denote by 'l'm the set of all functions 1p in m n vectors x ij , i = 1, ... , m, j = 1, ... , n, of
an n-dimensional space L satisfying the following conditions:
a) 1p is linear in every argument,
b) 1p is skew-symmetric as a function of x ioj ' j = 1, ... , n for any fIXed io,
c) 1p is symmetric as a function of xijo' i = 1, ... , m, for any given jo. Suppose that the
characteristic of k is greater than m. Show that every function 1p E 'l'm is given by its values
1pY''''Yn on the vectors xij = Yj and that 1pY''''Yn = dm 1pe, ... e n' where d is the determinant of the
coordinates of the vectors YI' ... , Y. in the basis e l ... en- Let ~ I' ... , ¢. E L*. The function 1p
for which 1pY""Yn = (det(¢,(y)))m is denoted by (¢I /\ ... /\ ¢.)m. Show that the space 'l'm is
1-dimensional and that (¢ I /\ ... /\ ¢.)m is a basis of it.
7. Generalize the construction of regular and rational n-dimensional differential forms
on n-dimensional varieties, replacing everywhere the space A'e~ by the corresponding
* Generalizations ofresults described here, to surfaces over fields of positive characteristic,
have since been proved by Bombieri and Mumford. See E. Bombieri and D. Husemoller:
"Classification and embeddings of surfaces". Algebraic Geometry. Azcata 1974. Proceedings
of Symposia in Pure Math. XXIX. (Footnote to new printing).
space 'Pm. The corresponding object is called a differential form of weight m. Show that in
the analogue to formula (6) of § 4.3 J must be replaced by J m• Show that a differential form
of weight m has a divisor, and that all these divisors form a single class, which coincides
with mKx . Generalize Theorem 2.
8. Determine the space of regular differential forms of weight 2 on a hyperelliptic curve.
Hint: Write them in the form f(dx)2jy2.
9. Verify the Riemann-Roch theorem for the case X = ]pl.
10. Let X be a smooth projective curve, g(X) = 1, and x E X. Show that for every n> 1
there exists a rational function Un on X for which vx(U.) = -n, v,(u.);;;.O for y#x. Use
Corollary 2 in § 5.6.
11. Under the conditions of Exercise 10 show that k(X) = k(u 2 , U3), where U2 and U3
are connected by a relationship of degree 3. Hint: To derive this relationship apply the
Riemann-Roch theorem to ft'(6x). Prove that [k(X) :k(u 2 , u 3 )] = 1, using Theorem 1 of§ 2.
Deduce that if the characteristic of the field is not 2 or 3, then any curve of genus 1 is
isomorphic to a curve given by the equation
where ft'(2x) = {1, u}, ft' (3 x) = {1, u, v}.

vr
12. Let Xl and X 2 be two curves of genus 1 given by the equations = U[ + aiu i + bi'
i = 1,2. Show that any isomorphism between them carrying the pole U l into U 2 is determined
by a linear transformation. Show that if the curves are isomorphic, then there exists an
isomorphism with this property. Hint: Use the structure of the algebraic group on Xl
and X 2 and the existence of shifts carrying a point into any other given point.
13. Under the conditions of Exercise 12 and assuming that bl # 0, b2 # 0, show that
two curves Xl and X 2 are isomorphic if and only if aVb~ = a~jbi.
14. Verify the relation degK=2g-2 for hyperelliptic curves and for smooth plane
curves.
15. Show that for a hyperelliptic curve the ratios of regular differential forms generate
a subfield of k(X) isomorphic to a field of rational functions. Starting out from this prove
that a smooth plane projective curve of degree m> 3 is not hyperelliptic.
16. Show that for a hyperelliptic curve the rational mapping corresponding to the
canonical class is not an isomorphism.
17. Show that if the mapping corresponding to the canonical class of a curve X is not
an isomorphism, then X is rational or hyperelliptic. Hint: If one of the conditions (2) in
§ 5.6 is not satisfied, then the Riemann-Roch theorem gives I(x);;;. 2 or I(x + y);;;. 2.
18. Show that a smooth hypersurface X of degree 3 in ]p4 is unirational. Hint: Using
Theorem 10 of Ch. I, § 6, show that X contains a line I. Using Exercise 8 in § 4 show that
there exists an open set U C X, Un 1# 0 such that the tangent fibre to U is isomorphic to
U x JA..3. Denote by ]p2 the projective space consisting of the lines that pass through the
origm of coordinates in JA..3. For a point ~ = (u, IX), UE In U, IX E ]p2, denote by qJ(~) the point
a
of intersection ofthe line lying in (Ju.x with X. Show that qJ determines the rational mapping
]pI x ]p2-+X.
19. Let 0 be a point of an algebraic curve X of genus g. Show (using the Riemann-Roch
theorem) that any divisor D with degD = 0 is equivalent to a divisor of the form Do - go,
where Do> 0, degDo = 9 (generalization of Theorem 5 in § 2).
20. Let XC]p2 be a smooth plane irreducible curve with the equation F = 0, let
a = (1X0: IXI : 1(2) ¢ X, x E X. The multiplicity Cx with which x occurs in a divisor of the form
ito exi : : is called the contact mUltiplicity at x. Show that Cx = ex is the ramification
mUltiplicity of the projection maping qJ: X -+]Pl from ex. Deduce that the number c = L Cx
xeX
- the number of tangents taken with the relevant mUltiplicities passing through ex - does
not depend on ex. It is called the class of the curve. Show that c= n(n- 1), n= degX.
Chapter IV. Intersection Indices
§ 1. Definition and Basic Propertie~
1. Definition of an Intersection Index. The theorems on the dimension

of an intersection of varieties, which we have proved in Ch. I, often allow
us to assert that certain systems of equations have solutions. However,
they say nothing about the number of solutions, assuming it to be finite.
The difference is like that between the theorem on the existence of roots of
a polynomial and the theorem that the total number of roots of a poly-
nomial is equal to its degree. The latter theorem is true only if we count
each root with its multiplicity. Similarly, to formulate general theorems
on the number of points of intersection of subvarieties, we ought to
assign certain multiplicities to these points. This will be done in the
present subsection.
We consider the intersection of subvarieties of codimension 1 on a
smooth variety X. We are interested in the case when the number of
points of intersection is finite. If dimX = n, and if C l , ... , Cl are sub-
varieties of codimension 1 with a non-empty intersection, then by the
theorem on the dimension of an intersection dim(C l n···nCz) > if
I < n. Therefore we naturally consider the case 1= n. The theory to be
°
applied later becomes simpler if instead of subvarieties of codimension 1
we consider arbitrary divisors. Thus, we consider n divisors Dl , ... , Dn
on a n-dimensional variety X. If x E X, X E nSUppDi and dimxnSuppDi
= 0, then we say that Dl , ... , Dn are in general position at x. This means
thatnSuppDi in some neighbourhood ofx consistSDfx only. If Dl , ... , Dn
are in general position at all points of the subvariety nSUppDi, this
subvariety consists of finitely many points or is empty. We then say that
Dl , ... , Dn are in general position.
We begin by defining the intersection index for effective divisors in
general position. Let Dl , .•. , Dn be effective divisors in general position at
a point x, having local equations fl' ... ,fn in some neighbourhood of
this point. Then there exists a neighbourhood U of x in which fl' ... , fn
are regular and do not vanish anywhere except at x. From Hilbert's
Nullstellensatz it foIIows that the ideal generated by the functions fl, ... ,fn
in the local ring (!)x of x contains some power of the maximal ideal mx of
this ring. Let
(1)
We consider the factor space (!)x/(fl, ... j",J (over k). Its dimension over
k is finite. To see this it is sufficient, by (1), to show that dimk(!)x/m~< 00.
The latter follows at once from the theorem on expansion in power
series: dimk(!)Jm~ coincides with the dimension of the space of poly-
nomials of degree less than I in n variables.
Henceforth we denote the dimension of a vector space E over k by
the symool I(E).
Definition 1. If Dl , .•. , Dn are effective divisors on an n-dimensional
variety X in general position at a point x E X, having local equations
fl' ··.,fn in some neighbourhood of this point, then the number
(2)
is called the intersection index (or multiplicity) of Dl , ... , Dn at this point
and is denoted by (Dl , •.. , Dn)x.
Actually, the number (2) depends only on the divisors Dl , .•. ,Dn and
not on the choice of their local equations fl' ... ,fn: if f1' ... ,f~ are other
local equations, then fi = figi' where gi, gi 1 E (!)x, therefore (fl' ... ,f,J
=(f1' ... ,f~).
Suppose now that Dl , ..• , Dn are divisors, but not necessarily effective.
We represent them in the form Di = Di - Di', Di ;;:. 0, Dt;;;:. 0, where the
divisors D; and Vi' do not have common components. This representation
is unique. Assuming that Dl , ... , Dn are in general position at x, then for
any permutation i l , ... , in and any I the divisors Di " ... , Di" D;; + I' ••• , D;:
are in general position at x, because SUPpDi = SuppD;uSuppD;'.
We now define the intersection index of Dl , ... , Dn at x by additivity,
that is, we set
(D1'···' D)
nx
= '\'
L.J '\'
i.J (- 1)n-I(D~ I], ...
D~Il' D~''1+1,"', D'!)
lnX·
(3)
ih···,i n O<'l<'n
Definition 2. If Dl , ... , Dn are divisors in general position on an

n-dimensional variety X, then the number
xenSuppD,
is called their intersection index and is denoted by (Dl , ... , Dn).

One could extend the sum formally to all points x E X, but only the
non-zero terms are written down above.
Note. The intersection index could also be defined without the
condition that X should be a smooth variety; however, then we have to
184 Chapter IV. Intersection Indices
confine ourselves to locally principal divisors. All our defmitions remain

valid.
Now we look at some examples, with the object of showing that our
definition of intersection multiplicity agrees with the geometric intuition.
Example 1. Let dim X = 1, t a local parameter at x, f the local equation
of the divisor D, vA!) = vAD) = m. Then (D)x = l(@x/(J))= l(@x/(t"'»)= m.
Thus, in this case the index (D)x is equal to the multiplicity with which x
occurs in D.
In the next example we assume that the Di are prime divisors, that is,
irreducible subvarieties of codimension 1.
Example 2. If x E Dl n· .. n Dm then according to the defmition
(Dl' ... , Dn)x:;;' 1. Let us clarify when (Dl' ... , Dn)x = 1.
Since fi E mx, hence (fl' ... ,fn) C mx, and l(@jm x) = 1, the condition
(Dl' ... , Dn)x = 1 is equivalent to (fl' ... , f J = mx· In other words, fl'" .In
must form a system of local parameters. In Ch. II, § 2.1, we have seen
that this holds if and only if the subvarieties Dl , ... , Dn intersect at x
transversally, so that x is a simple point on all the Di and neX,Di = x.
Example 3. Let dim X = 2, let x be a simple point on two curves Dl
and D2. According to Example 2, (Dl' D2)x > 1 if and only if the lines
eX,DI and e X,D2 are identical. Let fi(i = 1,2) be local equations of the
curves Db and let U and v be local parameters at x and fi=: (ctiu+ !3iv)(m;)
(i= 1, 2). Then the equations of the lines eX,Di are of the form cti~ + !3il1 =0
(i= 1,2) where ~=dxU,I1=dxv are coordinates on ex,x. Therefore
e X,DI=e x ,D2 if and only if ct2U+!32V=y(ctlU+!31v) for some YEk,
Y =I 0, in other words, f2 =: Y fl (m;). It is therefore natural to define the
order of contact of the curves Dl and D2 at x as the number I such that
there exists an invertible element g, g-l E@x, for which f2 =: gfl (m~+ 1),
and that such a g does not exist for larger values of the exponent I. We
show that the intersection index (D l , D2 )x exceeds by 1 the order of
contact of the curves Dl and D2 at x.
For this purpose we note that since x is a simple point on D l , we may
assume that fl is one ofthe elements of the system oflocal parameters at x.
On the other hand, g-l f2 is a local equation of D2. Therefore we may
assume that u and v are local parameters, that a local equation of Dl is u,
and one of D2 isf, withf =: u(m~+ 1). Thenf =:u+ cp(u, v) (m~+2), where cp
is a form of degree I + 1. Here cp is not divisible by u, otherwise Dl and D2
would have contact of order > 1. Therefore
cp(O,V)=CV I +l , C~O. (4)
According to the defmition of the intersection index
Evidently (fJx/(u) = @is the local ring of x on D1 , and the homomorphism

(fJx --+@ is the restriction of a function on X to the curve D1 . Furthermore,
(u,J)/(u) = (]), where Jis the image off in @.SinceJ E m~+ 1,] == iP(m~+ 2),
and since by (4) iP ¢m~+2 in (§, we have vA]) = 1+ 1 and l({§/(J)) = 1+ 1.
Thus, (D 1 , D2 )x = 1+ 1.
Example 4. Again, let dimX = 2 and let x be a singular point on D.
This means that f Em;, where f is a local equation of D. Therefore it is
natural to define the multiplicity of the singular point as the largest m for
which f Em';. We show that for every curve D' in general position at x
together with D,
(D,D')x~m (5)
and that there exist curves for which (D, D')x = m.
Let f' be a local equation of D'. We denote the ring {fJx/m'; by @and
the image of f' in {§ by ]. Since f E m~, we have
(D, D')x = 1({fJ x/(f, f')) ~ 1({§/(J)) .
By the theorem on power series expansions the ring (§ is isomorphic
to k[u, vJ/(u, v)m. Therefore, as a vector space it is isomorphic to
the space of polynomials in u and v of degree < m and is of dimension
m(m+ 1)
1 + 2 + ... + m = . If f' E m~, f' ¢ m/ 1, then to the elements of
I
2
the ideal (J) there correspond polynomials of the form J. g, where g
ranges over all polynomials of degree < m - l. Therefore
l((]))~ 1 + ... +(m-l)= (m+1-d Hm -l) .
Since f' E m, we have 1~ 1 and therefore l({§/(J)) = l(@) -l((])) ~ m.
Now we show that equality in (5) can be attained.
Let f== <p(u, v) (m,;+I), where <p is a form of degree m. We consider a
linear form in u and v that does not divide <po By a linear transformation
of u and v we can achieve that it is u, that is, <p(O, v)"# O. For D' we take the
curve with the local equation u. Then (D, D')x = l({fJx/(u,f)), and as we
have seen in the discussion of Example 3, this number is equal to m.
2. Additivity of the Intersection Index
Theorem 1. If D1 , ••• , Dn - 1 , D~ and D1 , ••• , Dn- 1 , D~ are divisors in
general position at x, then
(D 1 , ••. , Dn- 1 , D~ + D~)x
(1)
= (Dl' ... , Dn - 1, D~)x + (Dl' ... , Dn- 1 , D~)x·
Proof. First of all, it is obvious that Theorem 1 need only be proved
for effective divisors D1 , ... ,Dn_l,D~,D~. We assume from now on that
the divisors are effective.
We denote local equations of the divisors Dl, ... ,Dn-l,D~, D; by

fl' ···,fn-l,f~,f~/, the ring (9J(fl' ... ,fn-l) by t§, and the images of f~
and f~' in itT by f and g. Then
(Dl' ... , Dn-l,D~ + D;>:" = l(itT/(f. g»),
(Dl' ... , Dn- 1 , D~)" = 1(itT/(f)) ,
(D 1 , ••• , Dn- 1 , V::)" = 1(itT/(g») .
Since the sequence
is exact, we have
1(itT/(f· g») = I (itT/(g») + I (g)j(f . g»). (2)
If g is not a divisor of zero in itT, then multiplication by g determines an

isomorphism itT/(f) ~ (g)/(f g) and
I (g)j(f . g») = I (itT/(f)) . (3)
Therefore, if we can show that g is not a divisor of zero in t§, then (1)
follows from (2) and (3).
A sequence of n elements fl' ... ,fn of the local ring (9" of a simple
point of an n-dimensional variety is called a regular sequence if the /; are
not divisors of zero in (9,,/(fl' ... ,fi _ 1) for i = 1, ... , n.
The arguments above show that Theorem 1 is a consequence of the
following proposition.
Lemma 1. If D1 , ••• , Dn are divisors in general position at a simple
point x, then their local equations fl' ... , fn form a regular sequence.
The proof of Lemma 1 in its turn requires a simple auxiliary pro-
position.
Lemma 2. Over a local ring, the property of being a regular sequence
is preserved under a permutation of the elements of the sequence.
Proof of Lemma 2. It is enough to show that under a permutation of
two adjacent termsfi,fi+l in a regular sequence we again obtain a regular
sequence. We set (fl' ... ,fi-l) = a, (9Ja = A, and we denote by a and b
the images of fi and fi + 1 in A. Everything reduces to a proof of Lemma 2
for the regular sequence a, b in A. We have to prove that 1) b is not a
divisor of zero in A and 2) that a is not a divisor of zero modb.
1) Let c b = O. We show that then
(4)
for all I. From the fact that A is Noetherian and from Theorem 5 in
Ch. II, § 2, it then follows that c = o.
§ 1. Dermition and Basic Properties 187
The relation (4) can be verified by induction. If c = C1ai, then

C 1 a l b= o. Since a, b is a regular sequence, a is not a divisor of zero, hence
c 1 b = O. Again from the simplicity of the sequence a, b it follows that
C1 E(a), that is, cE(arl.
2) Let xa= yb. From the regularity of the sequence a, b it follows
that y=az, ZEA hence x=zb.
This proves Lemma 2.
Proof of Lemma 1. The proof proceeds by induction on the dimension
n of the variety X. From the condition of the lemma and the theorem on
the dimension of an intersection it follows that dimASupp(fl)Il···
IlSUpp(fn-l)) = 1. Therefore we can fmd a function u such that u(x) = 0,
that x is a simple point on the subvariety V(u), and that (fl)' ... , (fn-1)' (u)
are divisors in general position at x. It is sufficient to take for u the
equation of the hyperplane passing through x and not containing ex,x
nor any component of the curve SUPP(f1)1l·· ·IlSUpp(fn-1). We consider
the restrictions ofthe functionsf1' ... ,fn-1 to V(u). Evidently they satisfy
all the conditions of Lemma 1, therefore by the inductive hypothesis form
a regular sequence on V(u). Since the local ring of x on V(u) is ofthe form
(!)x/(u), we see that u, fl' ... ,fn-1 is a regular sequence. It follows from
Lemma 2 that then the sequence f1' ... /n-l' U is also regular.
To prove that the sequence fl' ... ,fn-l,fn is regular it remains for
us to verify that fn is not a divisor of zero in (!)x/(fl' ... ,fn-l). From the
condition on the functions f1' ... ,fn it follows that in some neighbourhood
of x the equations f1 = ... = fn = 0 have no solution other than x. Hilbert's
Nullstellensatz therefore shows that (f1' ... ,fJ)m~ for some I. In
particular, U I E(f1, ... ,fJ, that is, d=afn((f1, ... ,fn-1)) for some
aE(!)x· -
If fn were a divisor of zero in (!)x/(f1' ... ,fn-l)' then it would follow
that also d, hence u, is a divisor of zero in this ring. But this contradicts
the fact, which we have proved, thatfl' ... ,fn-1' u is a regular sequence.
Lemma 1, and with it Theorem 1, are now proved.
3. Invariance under Equivalence. We proceed to the proof of the

main property of intersection indices, which lies at the basis of all their
applications.
Theorem 2. If X is a smooth projective variety and if both D1 , ••• ,Dn- 1 ,Dn
and D 1, ... ,Dn- 1, D~ are divisors in general position and Dn and D~ are
equivalent, then
(D1' ... , Dn- 1, Dn) = (D 1, ... , Dn- 1, D~). (1)
By the condition of the theorem Dn - D~ = (f), and (1) is equivalent
to the fact that
(2)
when D1 , ••• , Dn - 1, (f) are in general position.
Representing Dj , 1, i , n - 1, as differences of effective divisors we

see that it is sufficient to prove (2) for Di> 0, 1, i , n - 1. From now
on we assume this to be the case. The proof of Theorem 2 makes use of a
more general concept of intersection index than we have used so far.
Let D1 , ••• , D" I, n, be effective divisors on an n-dimensional smooth
variety X. We say that they are in general position if dim n SuppDj
i=l, .. ·,1
= n - I or nSuppDj is empty. Suppose that this property holds and that
n
i=l, ... ,1
SuppDj=UCj (3)
where Cj are irreducible varieties of dimension n -I.

Under these conditions we can assign to the components Cj the so-
called intersection multiplicities, which coincide with the intersection
indices when 1= n, and consequently Cj consists of a single point.
The dermition of intersection multiplicities uses certain more general
concepts, which we now introduce.
Definition 1. We consider an irreducible subvariety C of an irreducible
variety X and functions f E k(X) that are regular for at least one point
CE C (of course, then they are regular in a whole open subset of C). Such
functions form a ring (!)e, the so-called local ring of the irreducible sub-
variety C.
It is easy to see that the ring (!)e does not change when X is replaced
by an open subset of it whose intersection with C is not empty. Therefore
we may regard X as affine. In that case the ring (!)e is obtained by the
construction described in Ch. II, § 1.1: (!)e = k [X]I-" where V is the ideal
of the subvariety C. In particular, (!)e is a Noetherian local ring. We
denote its maximal ideal by me. Every function f E (!)e determines by
restriction a rational function on C. From this it is easy to deduce that
(!)e/me = k(C).
Next we give some properties of local rings at prime ideals whose
verification is obvious. We recall that in Ch. II, § 1.1, we have dermed a
homomorphism cp: A -+ Ai->' To every ideal a C A there corresponds an
ideal cp( a) C Ai->' generated by the elements cp(x), x Ea. It consists of the
pairs (a, b) for which there exists a b' ¢ V such that ab' Ea.
An immediate verification shows that cp(a) = Ai-> when a ct. V, and if
aCV, then
(4)
where p is the image of V in A/a.
The latter property is equally easy to verify: let a and b be ideals
of A, a) b, and suppose that there is an isomorphism of A-modules
alb ~ A/q, where q is a prime ideal in A. Then under localization with
respect to a prime ideal peA we have cp(a») cp(b) and

cp(a) = cp(b) , when q<tp,
(5)
cp(a)/cp(b)~Ap/cp(p), when q=p.
The other concept required to defme intersection multiplicities is the

length of a module.
Deimition 2. A module M over a ring A is said to be a module of finite
length if it has a finite sequence of submodules
(6)
for which the factor modules MJMi + 1 are all simple, that is, do not
contain submodules other than zero and the whole module. From the
Jordan-Holder theorem it follows that all such chains consist of the
same number n of modules, which is called the length of the module and is
denoted by I(M).
If A is a field, then the concept of length becomes the dimension of a
vector space.
Later we shall use two almost obvious properties of this concept.
If a module has finite length, then the same is true for any of its
submodules and factor modules.
If a module M has a chain of submodules (6) in which the lengths of
the modules MJMi + 1 are finite, then the length of M is also fmite and
I(M) = I:. I (MJMi + 1)'
Finally we are in a position to proceed to the definition of inter-
section multiplicities. It is an exact copy of the defmition of an intersection
index. Let C be one of the components Cj in (3). We choose a point x E C
and local equations J; of the divisors Di in a neighbourhood of this point.
Then fi E (!Je> and the ideal a = (fl ' ... , it) c (!Je does not depend on the
choice oflocal equations nor of the point x. For if gl' ... , gl are other
local equations in a neighbourhood of another point, then hand gi are
local equations of the divisor Di on an entire open set intersecting C.
Hence it follows that figi 1 E (!Je and gdi 1 E (!Je> therefore
(f1' ... ,J;) = (gl' ... , gl)'
Lemma. The module {!}e/a is of finite length.
For since C is an irreducible component of the subvariety determined
by the equation f1 = ... = it = 0, there exists an open affme subset
U C X intersecting C in which these equations determine C. Then by
Hilbert's Nullstellensatz (f1"'" it) ) a'C for some r > O. Consider now
the local ring Ap , where A = k[U], p = ae. As we have already said,
Ap={!}e, CP((f1, ... ,it»)=a, and cp(ad=me' Therefore in {!}e we have
a) m~. To check that the module {!}c/a is of fmite length it is sufficient

to verify this for the module M = (!}c/m~.
Considering the sequence of submodules M j = m~/m~ we see that
it is sufficient to check that the modules m~/m~+ 1 are of fmite length.
But under the action of A on this module the ideal mc annihilates all the
elements. Therefore A/mc = k(C) acts on the module, so that m~/m~+ 1
is a vector space over the field k( C), and its length is the same as its
dimension over this field. Since A is a Noetherian ring, this module is
finitely generated, hence is a finite-dimensional vector space, which
proves the lemma.
Definition 3. The number l({!}c/a) is called the intersection multiplicity
of the divisors D1, ... , Dl in the component C. It is denoted by (D 1, ... ,DI)c.
Theorem 2 is a simple consequence of two propositions, which we
now state.
Proposition 1. If D1, ... , Dn are divisors in general position at x and
D1 ;;;. 0, ... , Dn- 1 ;;;. 0, then
r
(Dl,···,Dn)x= L
j= 1
(D1,···,Dn-1)dQciDn)}x, (7)
where C 1, ... Cr are all the irreducible components of the variety

SUPpDl n ... n SuppDn _ 1 , and QdDJ is the restriction of Dn to Cj (see
Ch. III, § 1.2).
Note that since D1 , ... , Dn are divisors in general position at x, we
have dimCj = 1, XE SUPPQciDn), and the intersection index (Qc)Dn)}x is
defmed (on the curve CJ
Proposition 2. For a curve C and a locally smooth divisor D on it
(D)x= L (v*(D))y, (8)

v(y)=x
where v: cv --? C is the normalization of C.

We derive at once Theorem 2 from these propositions and postpone
their proof to the next subsection.
We write the intersection index in the form
(D 1, ... , Dn) = L (D 1, ... , Dn)x

XEX
According to Proposition 1
r
(D1,· .. ,Dn)= L (D1, ... ,Dn-1)cj L (QciDn)}x,

j= 1 XECj
and according to Proposition 2
If Dn is a principal divisor, Dn = (f), then so is the divisor

(v* Qc) (Dn): (v* Qc) (Dn) = (g) and (g))y = vig)·
Since X is a projective variety, the curves Cj are projective, and by
Theorem 10 of Ch. II, § 5, so are the Cj. According to the corollary to
Theorem 1 of Ch. III, § 2, L
Vy(g) = deg(g)) = 0, from which it follows
yeCj
that (Dl' ... , D n - 1, (f)) =0.
4. End of the Proof of Invariance. Now we prove Proposition 1.

Let f1' ... ,fn-1 be local equations of the divisors D 1 , ... ,Dn- 1,
a = (f1, ... , fn-1) C @x, @x/a = i§x, J the image of f (the local equation of
Dn) in i§x' The definition of the intersection index shows that
(1)
and Lemma 1 of § 1.2 asserts that J is not a divisor of zero in i§x'

First of all we have to clarify what are the prime ideals of i§". We
denote by Vi the collection of functions in @x that vanish identically on
C i, and by 1.\ the image of Vi in i§". Evidently,
(2)
is the local ring of x on Ci . We denote by itt the maximal ideal of i§", the
image of the ideal mx C @x'
Lemma 1. The ideals P1, ... , Pr and itt are the only prime ideals of i§".
The assertion of the lemma is equivalent to the fact that V1, ... , Vr
and mx are the only prime ideals of@x containing a. Let a eve @x, V being
a prime ideal. We consider an affine neighbourhood U of x in which
f1' ... ,fn-1 are regular, and we set A = k [U], ~ = An V. Clearly ~ is a
prime ideal. We denote by V the subvariety which it defines in U. Since
V) a, we have V C C 1 U·· . U C" and since ~ is prime, V is irreducible.
Therefore V either coincides with one of the Ci, and then ~ = A n Vi' or
else V is a point y E U (we recall that the Ci are one-dimensional). In the
latter case, if y # x, then ~ and hence also V contains a function that does
not vanish at x. Since @x is a local ring, we would then have V= @x,
(whereas the ring itself is not considered as one of its prime ideals.)
Thus, the only remaining possibility is that ~ = Anm x . Since V= ~. @x,
it follows easily that V= Vi for some i, i = i, ... , r, or V= m x , as the lemma
claims.
°
Lemma 2. Every Noetherian ring A has a sequence of ideals
A = q I ) q z ) ... ) qs = such that
(3)
where P is some prime ideal of A (:p depending on i).
Proof. We consider an arbitrary element a E A, a =1= 0, and denote by
Ann(a) the annihilator of a, that is, the set of all x E A for which xa = 0.
Since A is Noetherian, any sequence Ann (a) C Ann (a l ) C Ann (a z) C ...
breaks off, therefore we may assume that already a has the following
property: from Ann (a) C Ann (a'), a' =1= 0, it follows that Ann (a) = Ann (a').
Let us show that then the ideal Ann(a) is prime. For if bc E Ann (a),
b¢Ann(a), then abc=O, ab=l=O and therefore Ann(a)CAnn(ab), hence
Ann(a)=Ann(ab), by the property of a. But C E Ann (a b), hence CE Ann(a).
This shows that the ideal Ann(a) is prime. We set Ann (a) = p. The
homomorphism x -+ a x determine an isomorphism of modules
(a):::::: AlP.
It now remains to go over to the ring Al = A/(a) and to apply the
same argument to it. So we obtain an ascending sequence of ideals
qs C qs+ I C "', and (3) holds for every pair of adjacent ideals. Since A is
Noetherian, the sequence must break off.
According to Lemma 2, iff,. contains a chain of ideals qi having the
property (3).
Lemma 3. If kj is the number of times P= Pj occurs in (3), then
(DI' ... , Dn-1)ej = kj.
Proof. From the definition of a local ring and a prime ideal it follows
at once that (lDJn. TJ
= lD e J . Applying the relation (4) of § 1.3 we find that
lDe/(fl' ... ,fn-I) = (lDx/UI' ... ,fn-l))pj = (iff,.)Pj'
To the chain of ideals qi we apply the mapping cp corresponding to the
ideal Pj. So we obtain the chain
lDe/Ul' .. ·,fn-I) = CP(ql) ~ cp(qz) ~ ... ~ cp(qs) = 0.
The relation (5) of § 1.3 shows what the factors in this cha~ are: ifp =1= Pj'
then CP(qi)=CP(qi+l), and ifp=pj' then CP(qi)/CP(qi+I)= (lDxlP)pj= k(C).
Thus, the length of the module lDe/(fl' ... ,fn-I) is equal to kj' as Lemma 3
claims.
Now we can prove Proposition 1. We have the exact sequence
0-+ qz -+iff,.-+i!5jqz -+0.
Here two cases are possible: 1) the prime ideal P corresponding to qz
by virtue of (3) is equal to in, 2) P = pj for some j = 1, ... , r.
In case 1) i!5jqz:::::: iff,./in = k. Since J is not a divisor of zero in iff,., the
mapping a -+J. a carries any ideal in @x into an ideal that is isomorphic
to it as an ~-module. In particular,
iff)q2c::;j~/fq2 . (4)
From the diagram
we find that
I (iff) jq2) = l(~/ j~) + I (jiff)lPz) = l(iffx/q2)+ l(q2!lq2)
and all these numbers are finite. By (4) it follows that I (iff) j~) = l(q2/ j q2)'
In case 2) ~q2 c::; (!}x,Cj' and we have the sequence
0-+q2/ jq2 -+~/ j~-+{!}x,c/ j{!}x,Cj-+O. (5)
This sequence is exact. The verification is completely obvious except
at one place: namely, the fact that the homomorphism q2/ j q2 -+ iff) j~
is an embedding. This follows at once from the fact that the image 1 is
not a divisor ofze~o in the ring iff)q2 c::; (!}x,cj" For this ring has no divisors
of zero at all, and f is not equal to zero in it, because f #- 0 on Cj • From (5)
we have
I (iff) j~) = l(q2/ j q2) + I ({!}x, c/ j{!}x, c) = l(q2/ j q2) + l({!}x,c/(j)).
Repeating the same argument s times we obtain the formula
(Dl' ... , Dn)x = "kk)({!}x,c/(j)) = "kkAed(f)))x'
Here lej is the number of indices t <,s for which in the sequence
iffx = q 1 ) ... ) qs = 0 we have qt/qt+ 1 c::; iffx/pj' Lemma 3 guarantees that
this number is ,equal to (Dl' ... , Dn-1)cj'
Proof of Proposition 2. We denote by Yl"'" Yl inverse images of x
under the nor!llalization mapping v.:, C -+ C, by (!}Yi their local rings on C,
and we set {!} = n{!}Yi' Obviously (!}) (!}x [if we regard k(C) and (!}x as
embedded in the field k(C) by means of the mapping v*]. We need the
following property of these rings:
Lemma 4. There exists an element dE {!}x, d #- 0, such that
(6)
We consider an affme neighbourhood U of x and its normalization
V, and we set A = k [UJ, B = k [V]. One of the main steps in the proof
of the theorem on the existence of a normalization of an affme variety
consists in establishing that B is a module of fmite type over A. Hence
and from the fact that B is contained in the field of fractions of A it

follows that
d·BCA (7)
for some dE A, d i= O. It is easy to see that
&= B· (!)x' (8)
For evidently Be &, therefore B(!)x C &. On the other hand, the lemma in
Ch. III, § 2.1, shows that & C B(!)x, and (6) follows at once from (7) and (8).
Now it is not difficult to conclude the proof of Proposition 2. By
Lemma 4, I(&/(!)J.;;;; 1(&/d &) and by Theorem 3 of Ch. III, § 2, I(&/d&)
= L I«(!)y/d(()y) = L. vy,(d), therefore 1(&/(!) x) < 00. From the diagram
y,
it follows that
I(&/! (!)x) = I(&/(!)x) + I«(!)xI! (!)x) = 1(&/! &) + 1(.f &/! (!)x),
and all these number are finite. Since! is not a divisor of zero in &, we
have 1(&/(!)x)=I(.f&/!(!)x), from which we obtain that I«(!)x/f(!)x)
= I (&/.f&x)' Finally, Theorem 2 ofCh. III, § 1, yields I(&/f&) = vy,(.f) L
y,
= L (v*(.f))y,. Since I«(!)x/! (!)x) = (D)x, this proves Proposition 2.
y,
5. General Definition of the Intersection Index. Theorem 2 and the

theorem on shifting the support of a divisor away from a point
(Theorem 1 of Ch. III, § 1) allow us to defme the intersection index of any
n divisors on an n-dimensional smooth projective variety, without any
restrictions of the type of general position.
For this purpose we need two lemmas.
Lemma 1. For any n divisors D1 , ..• , Dn on an n-dimensional variety X
there are n divisors D~, ... , D;' such that Di ~ D; (i = 1, ... , n) and D~, ... , D;'
are in general position.
Suppose that we have already found such divisors D~, ... , D; such
that Di ~ D[ (i = 1, ... , Qand that dim (Supp D[ n··· nSuppD;) = n-I, or
that this intersection is empty. Let
SuppD~n··· nSuppD; = C 1 u··· uCr
§ 1. Defmition and Basic Properties 195
be a decomposition into irreducible components. On each of the com-

ponents Cj we chose one point Xj' and by using the theorem on shifting
the support of a divisor we find a divisor D;+ 1 such that D;+ 1 ~ D 1+ 1
and XjESuppD/+ 1(j= 1, ... , r). Then SuppD/+ 1 a fortiori does not contain
any of the components C j' and by the theorem of the dimension of an
intersection
dim(SuppDlfl'" flSuppD,+1)=n-l-1,
if this intersection is not empty. When we arrive at 1= n in this manner,

we have obtained the required system of divisors.
Lemma 2. If the divisors D 1, ... , Dn and D 1, ... , D~ are in general
if Di ~ Di (i = 1, ... , n), then
position and
(1)
If D1 = D 1, ... , Dn _ 1 = D~ _ l' this is the assertion of Theorem 2. Let

us show that (1) is true if D1 =D1, ... ,Dn-l=D~-I' For I=n we then
obtain our proposition.
We use induction on l. Suppose that the assertion is true for values
smaller than I. Since both system of divisors D1, ... , Dn and D 1, ... , D~
are in general position, we have dim Y = dim Y' = 1, where
Y= n
i*n-l+1
SUPpDi' Y' = n
i*n-I+1
SuppDi. On each component of each
of the varieties Y and Y' we choose one point, and according to the
theorem on the shift of the support of a divisor we find a divisor D;-l+ 1
such that S upp D;-I + 1 does not pass through any of these points and that
D;-1+1 ~Dn-l+1' Then both systems D 1, ... ,Dn- 1, D;-1+1, ... ,Dn and
D 1, ... , D~-l' D;-l+ l' ... , D~ are in general position.
By Theorem 2
(D1' ... , Dn) = (D1' ... , Dn-" D;-l+ 1> ••• , Dn),
(2)
(Dl' ... , D~) = (Dl' ... , D~_" D;-l+ 1> ••• , D~) .
The right-hand sides of (2) are equal to each other by the inductive
hypothesis (they have already n -1 + 1 equal divisors), and this proves
Lemma 2.
U sing Lemmas 1 and 2 we can define the intersection index (D1' ... , Dn)
for any n divisors on a smooth n-dimensional variety, without requiring
that they are in general position. For this purpose we fmd arbitrary
divisors D1, ... , D~ satisfying the conditions of Lemma 1, so that the index
(Dl' ... , DJ is defined, and we define (D1' ... , Dn) by the equality (D1' ... , Dn)
= (Dl' ... ,D~).
We have to show that this defmition does not depend on the choice
of the auxiliary divisors Di, ... , D~, but precisely this is guaranteed by
Lemma 2.
We can now talk, for instance, of the sef-intersection (C, C) for a
curve C on a surface X. This number is also denoted by (C 2 ). We give
some examples how it can be computed.
Example 1. X = ]p2, C a straight line. By definition, (C 2) = (C', C"),
where C' '" C" '" C and C' and C" are in general position. For C' and C'"
we may take, for example, two distinct lines. They intersect in a unique
point x, and (C',C")=(C',C")x= 1, because they are transversal at this
point. Therefore (C 2 ) = 1.
Example 2. X is a smooth surface in ]p3. Let X be given by the
equation F(xo: Xl: X 2 : X3) = 0, degF = m.
We calculate (E2), where E is the divisor of a plane section (in Ch. III,
§ 1.2, we have shown that the divisors of all plane sections are equivalent).
By defmition (E2) = (E', E"), where E' '" E" '" E and E' and E" are in
general position. For E' and E" we choose distinct plane sections. Then
E' = 'EliC i, where Ci are plane curves and 'Eli deg Ci = m (we simply have
to substitute in F the equation of E' and decompose the resulting form in
three variables into irreducible factors). We choose E" so that it is
transversal to all the Ci ; that such a choice is possible follows easily from
a count of the dimension of the set of non-transversal planes. Then, by
Bezout's theorem in Ch. III, § 2.2
(E', E") = 'E li(C i, E") = 'E Ii degCi = m= degX .
Therefore (E2) = degX.

Example 3. Suppose that on the surface X of Example 2 there lies a
line L. We wish to compute (L2 ).
We draw a plane through L and denote by E the corresponding plane
section. Then L is contained as a component in E:
First we compute (C 2 ). To do this we observe that at the points of

intersection of Land C the curve E has a singular point, and this means
that the plane containing it coincides with the tangent plane to X at this
point. We consider another plane passing through L, but distinct from
the tangent planes to X at the points of LnC. This plane determines a
divisor E'=L+C', and the points LnC and LnC' are all distinct.
This means that C n C' == 0 and (C 2 ) = (C, C) = O. So we have obtained
the equations
m= (E2) = (E, L+ C)= (E, L)+ (E, C)= 1 + (E, C),
(C,E)=m-1,
m- 1 = (E, C)= (L, C)+ (C 2 ) = (L, C),
1 = (E, L) = (If) + (L, C) = (L2) + m - 1 ,
(L2) = 2- m.
Observe that (L2) < 0 for m> 2. In fact, lines can lie on surfaces of any
degree, for example, the line Xo = Xl' X 2 = X3 on the surface xO - xT
+ xi' - x'3 = o.
Exercises
1. Let X be a surface, x a simple point on it, u and v local parameters at x, and f the
local equation of a curve C in a neighbourhood of x. Iff = (au + bv) (cu + dv) + g, g E m~,
and if the linear forms au + b v and c u + d v are not proportional, then x is called a double
point of C with distinct tangents, and the lines of ex with the equations au + b v = 0 and
cu + dv = 0 are called the tangents at x. Under these assumptions let C' be a smooth curve
on X passing through x. Show that (C, C')x > 2 if and only if ex.C' is one of the tangents
to C at x.
2. Let C = V(F), D = V(G) be two plane curves in JA2, and x a simple point on each
of them. Let f be the restriction of the polynomial F to the curve D, and vx(f) the order of
the zero of this function at x on D. Show that this number does not change when F and G
interchange places.
3. Let Y be a smooth irreducible subvariety of codimension 1 of an n-dimensional
smooth variety X. Show that for divisors D " ... , Dn _ I in general position with Y at x,
(DI' ... , Dn- Y)x = (Qy{DI)' ... , Qy{Dn- d)x, the second intersection index being computed
"
on Y.
4. Find the degree of the surface vmOp2) (vrn is the Veronese mapping).
5. Let X be a smooth projective surface contained in the space lPn, and L a projective
subspace of lPn of dimension n - 2. Suppose that L and X intersect in finitely many points,
and that at I of these points the tangent plane to X intersects L in a line. Show that the
number of points of intersection of X and L does not exceed degX -I.
6. The same as in Exercise 5, but the dimension of L is n - m, m;;. 2. Show that the
number of points of intersection of X and L does not exceed deg X -I - m + 2. Hint:
Draw through L a suitable linear subspace satisfying the conditions of Exercise 5.
7. Show that (Hn) = degX for an n-dimensional smooth projective variety X C lPN,
where H is the divisor of the intersection of X with a hyperplane in lPN
8. Show that if DI , ... , Dn - I are effective divisors on an n-dimensional variety in general
position and if C is an irreducible component of the intersection of their supports, then
(D " ... , Dn _ de = min (D I , ... , Dn _ I' D)x> where the minimum is taken over all points x E C
and all effective divisors D for which XES upp D.
9. Calculate (D I ,D2)c, where DI and D2 are given in JA3 by the equations x=o and
x 2 + y2 + xZ= 0, and C is the line X= 0, y= O.
§ 2. Applications and Generalizations of Intersection Indices

1. Bezout's Theorem in a Projective Space and Products of Projective
Spaces. Theorems 1 and 2 of § 1 put us in a position to calculate inter-
section indices of any divisors on a variety X, if only the group Cl(X) is
sufficiently well known to us. We show in this two examples.
Example 1. X = ]pn. We know that Cl (X) ~ Z, and for a generator
of this group we can take the divisor E of a hyperplane. Every effective
divisor D is a divisor of a form F, and if degF = m, then D ~ mE. Hence it
follows that if Di",miE (i= 1, ... ,n), then
(1)
because evidently (En) = 1.

If Di are effective divisors, that is, correspond to forms Fi of degree m i
and are in general position, then the points of the set nSUppDi coincide
with the non-zero solutions of the system of equations
Fn(xO ... xn) = 0 .
For such a point (or solution) x the index (Dl' ... , Dm)x is naturally called
the multiplicity of the solution. Then the Eq. (1) shows that the number
of solutions of a system of n homogeneous equations in n + 1 unknowns
is either infinite or equal to the product ofthe degrees, provided that their
solutions are counted with their multiplicities. Here only non-zero
solutions are considered, and proportional solutions are counted as one.
This result is called Bezout's theorem in the projective space ]pn.
Example 2. X =]pn X ]pm. In this case Cl(X) = Z EEl Z. Every effective
divisor D is determined by a polynomial G homogeneous with respect
to the variables x o, ... , Xn (coordinates in ]pn) and Yo, ... , Ym (coordinates
in ]pm). If G is of degree of homogeneity k and 1, then D ~ (k, 0 determines
an isomorphism Cl(X) ~ Z EEl Z. In particular, as generators of Cl(X)
we can take a divisor E determined by linear forms in the Xi' and a divisor
F determined by linear forms in the Yi' Then D '" kE + IF.
Let Di ",kiE + liF (i = 1, ... , n + m). Then
(D 1 , .. ·, Dn+J = 'Lki1 .. · ki)it ... ljs(E, ... , E, F, ... , F),
'---v-' '---v-'
where the summation extends over all permutations (i 1 ... ir jl ... js) of
the numbers 1, 2, ... , n + m for which i 1 < i2 < ... < ir;jl <j2 < ... <js'
§ 2. Applications and Generalizations of Intersection Indices 199
Let us calculate the intersection index

(E, ... ,E,F, ... ,F). (2)
'-yo-' '-yo-'
If r > n, then we can fmd r linear forms E l' ... , E, without common zeros,
and therefore
(E, ... ,E,F, ... , F)=(E 1, ... , E"F, ... ,F)= o.
'-yo-'
The matter is similar if s> m. Since r + s = n + m, the index (2) can be

different from zero only for r = n, s = m. In this case we can take for
E l' ... , Em F 1, ..• , Fm the divisors determined by the forms
Xl' •.• ,Xno Y1, ... ,Ym. These divisors have a unique point in common
(1 :0: ... : 0; 1: 0 ... : 0). They intersect transversally at it, as is easy to verify
by going over to the open set X o =F 0, Yo =F 0, which is isomorphic to the
affine space JAn+m. Thus,
(k1E + 11F, ···,k,,+mE + In+mF) ='i:.k;1 ... kinliI ... lim' (3)
where the sum extends over all permutations (i1 ... i,J1 ... jJ of the
numbers 1,2, ... , n + m in which i1 < i2 < ... < in; j1 <j2 < ... <jm. This
proposition is called Bezout's theorem in the variety lPn x lPm.
A common feature of the examples we have analysed is the fact their
groups CI(X) are finitely generated. It is natural to ask whether this is
true for any smooth projective variety X. This is not so, and a counter-
example is given by a plane cubic curve for which CI(X)) ClO(X),
CI(X)/Clo(X) ~Z,and the elements of the groups ClO(X) are in one-to-one
correspondence with the points of X. Therefore, if k is the field of complex
numbers, then the group ClO(X) is even uncountable.
However, this "bad" subgroup ClO(X) has no influence at all on the
intersection index (D) = degD - it consists of divisors of degree o. A
similar situation arises in the case of arbitrary smooth projective variety
X. Namely, it can be shown that if a divisor D is algebraically equivalent
to zero (for the definition see Ch. III, § 3.5), then (D1' ... , Dn - 1, D) = 0 for
arbitrary divisors D1 , ••• , Dn -1. Thus, the intersection indices depend
only on the elements of the group Div(X)/Diva(x). For this group
Theorem D of Ch. III, § 3.5, asserts that it is always fmitely generated.
Obviously, if E 1 , .•• , E, are generators of this group, then to know
arbitrary intersection indices of divisors on X it is sufficient to know
finitely many numbers (E;I' ... , E;J, just as we have seen in Example 1
and 2. In other words, an analogue to Bezout's theorem holds on X.
2. Varieties over the Field of Real Numbers. The various versions

of Bezout's theorem proved above have some pretty applications to
algebraic geometry over the field of real numbers.
200 Chapter IV. "Intersection Indices
We return to Example 1 of § 2.1 and assume that the equations

Fj = 0 (i = 1, ... , n) have real coefficients; we are interested in real solutions.
If degFj = mj and the divisors Dj are in general position, then
(Dl' ... , Dn)=m 1 ••• mn , as was shown in § 2.1. According to the defmition,
(Dl' ... , Dn) = ~ (D 1 , ••• , Dn)x, where the sum extends over the solutions x
of the system Fl = 0, ... , Fn = O. Of course, here we must consider both
real and complex solutions. However, since the polynomials Fj have
real coefficients, together with any complex solution x, the system also
has the conjugate complex solution x. From the definition of the inter-
section index it follows at once that (Dl' ... , Dn)x = (Dl' ... , Dn)x, therefore
(Dl' ... ,Dn)=~(Dl' ... ,Dn)y (mod 2), where now the sum extends only
over the real solutions. In particular, if (Dl' ... , DJ is odd (and this is
equivalent to the fact that all the mj = degFj are odd), we see that there
exists at least one real solution. This proposition has been proved under
the assumption that the divisors Dj are in general position. However, the
following simple argument allows us to get rid of this restriction.
The fact is that the theorem on shifting the support of a divisor has
in our case a perfectly simple proof and in a more explicit form. Namely,
we can take a linear form I that is different from zero at all the points
Xl' ... , X, from which we wish to shift the support of the divisor. If a
divisor D is determined by a form F of degree m, then the divisor D'
determined by the form Fe = F + 8l'" satisfies all the conditions of the
theorem if only F(x j ) + el(Xjr:;i: 0 (j = 1, ... , r). These conditions can be
satisfied for sufficiently small values of 8.
Now we show how to get rid of the restriction concerning general
position in the proposition we have proved above about the existence
of a real solution of a system of equation of odd degree. Let
(1)
be any such system. By what we have said above, we can find sufficiently
small values of 8 for which the divisors defined by the forms Fj • e = Fj + 8 l'{"
are in general position.
By what we have already proved, the system F1 •• = 0, ... , Fn.e= 0 has
a real solution X •• Since a projective space is compact, we can find a
sequence of numbers 8m --. 0 such that the points Xe m converge to a point
X E IPn. Since then Fj • em --. Fj , we see that X is a solution ofthe system (1).
Let us summarize what we have proved.
Theorem 1. A system of n homogeneous real equations in n + 1 un-
knowns has a non-zero real solution if the degrees of all the equations are
odd.
Very similar arguments apply to the variety IPn x IPm (see Example 2
of § 2.1). We obtain the following result.
Theorem 2. A system of real equations

Fi(XO: ... :xn;YO:'" :Ym)= O(i= 1, ... , n+ m)
has a non-zero real solution if the number L. kil ... k i)j, .. . ljm is odd.
Here Ii and lj are the degrees of homogeneity of a polynomial Fi in
the first and second system of variables, respectively, and a zero solution
is one for which Xo = ... = Xn = 0 or Yo = ... = Ym = O.
Theorem 2 has interesting applications in algebra. One of these
refers to the problem of division algebras over the field of real numberslR.
If the rank of such an algebra is n, then it has a basis e 1 , ... , en and is given
by a multiplication table
n
eiej = I cLez (i,j= 1, ... ,n). (2)
z= 1
We do not assume the algebra to be associative, therefore the cL can be
arbitrary. A division algebra is one for which an equation
ax=b (3)
is soluble for every a =f. 0 and every b. It is easy to see that this is equivalent
to the absence of divisors of zero in the algebra. To see this it is sufficient
to consider the linear transformation q>: q>(x) = ax in the vector space
formed by the elements of the algebra. Condition (3) indicates that the
image of q> is the whole space. This is equivalent to the fact that the kernel
of q> is zero. The latter condition means that there are no divisors of zero
in the algebra, that is, from x Y = 0 it follows that x = 0 or Y = O. If
n n
X= I xie i , Y= I yje j ,
i= 1 j= 1
then it follows from (2) that

n
zz= I c!j Xi Yj (/=1, ... ,n).
i,j= 1
Thus, division is possible in the algebra if the system of equations

n
Fz(X,y)= L cfjxiYj=O (l=1, ... ,n) (4)
i,j= 1
has no real solutions in which (x 1 , ... ,xn)=f.(0, ... ,0) and (Yl""'Y")
=f. (0, ... ,0). These equations almost fall under the conditions of
Theorem 2. The only difference is that the polynomials F z determine
equations in IPn-l x IPn-l, but their number n is not equal to the dimen-
sion 2n - 2 of this space. Therefore we choose an arbitrary integer
1 ~ r~n-1 and set Xr+2 = ... = xn=O, Yn-r+l = ... = Yn= O. The equa-
tions F 1(X 1, ... ,Xr+ 1,0, ... ,0;Y1, ... ,Yn-r+1'0, ... ,0)=0 (1=1, ... ,n) are
now given in lP r x IP"-r and a fortiori have no non-zero real solutions.
By Theorem 2 this is possible only if the sum
(5)
is even, and this must hold for all r = 1, ... , n - 1. In our case the forms
= 1; = 1 and the sum (5) is equal to the number
FI are bilinear, so that k;
of its terms, namely G). SO we see that the system (4) has no non-zero
real solutions only if all the numbers G) are even for 1, ... , r= n - 1.
This is possible only if n For our condition on G) can be expressed

= 21.
in the following way: in the field IFz of two elements (T + 1)" = T" + 1.
If n = 21 . m, where m is odd and m> 1, then in IF2
(T + 1fm= (T Z1 + 1)m= TZlm+m T Z' (m-1) + ... + 1 =F T"+ 1 .
We have proved the following-results:
Theorem 3. The rank of a division algebra over the field of real numbers
is a power of two.
lt can be shown that a division algebra exists only for n = 1, 2, 4, 8.
The proof of this fact uses rather delicate topological reasoning.
Applying similar arguments we can investigate for what values of m
and n the system of equations
m
L
i,j= 1
c!jXiYj=O (l=1, ... ,n)
does not have non-zero real solutions. This question is interesting in

that it is equivalent to the problem of ellipticity of the system of dif-
ferential equations
~
1..-
~
1..-
1
Cij::'l
OU j -0
- (i= 1, ... ,m).
1=1 j=l uX 1
3. The Genus of a Smooth Curve on a Surface. In the geometry on a

smooth projective surface X an important role is played by the following
formula, which expresses the genus of a smooth curve C C X in terms of
certain intersections indices:
_ (C,C +K) l'
gc- 2 +, (1)
here gc is the genus of C, and K is the canonical class of X.

This formula could be proved by using the tools already known to us.
However, a clearer and geometrically more lucid proof follows from the
simplest properties of vector bundles. This will be given in Ch. VI, § 1.4.
Here we only quote some of its applications.
1. If X = JP2, then Cl (X) = 7l, and a generator is the class L containing
all straight lines. If degC = n, then C", nL. Since K = - 3L and (L2) = 1,
n(n- 3) (n-1)(n- 2)
the formula (1) in this case gives g= 2 + 1= 2 .
This result was obtained by other means in Ch. III, § 5.4.
2. Let X be a smooth quadric in JP3. Let us clarify how to classify
smooth curves on X by their geometric properties.
An algebraic classification is perfectly clear. Since X ~ JP1 X JP l , any
curve on X is given by an equation F(X O:X1;YO:Yl)=O, where F is a
polynomial homogeneous in Xo and Xl as well as Yo and Yl' We denote
the degrees of homogeneity by m and n, respectively. The number of
coefficients of such a polynomial is (m + 1)(n + 1), hence all curves
given by equations of degree of homogeneity m and n correspond to
points of the projective space JP mn + m + n• Since for arbitrary positive m
and n there exist smooth irreducible curves, for example, the curve with
the equation
2x'3 YO + x'3,Vi + xi Yi = 0 ,
to smooth irreducible curves there correspond points of a non-empty
open subset of JP mn + m + n•
We have seen in § 2.1 that Cl (X) = 7l EB 7l, and if the curve C is given
by an equation with the degrees of homogeneity m and n, then
C "'mE+nF, (2)
where E and F are as in Ex. 2. Thus, curves corresponding to given
numbers m and n are effective divisors of the class mE + nF.
The classes E and F correspond to two families of rectilinear gene-
rators on X. It is easy to find the intersection indices of curves given in
the form (2): if
C '" mE + nF, C '" m' E + n' F , (3)
then
(C,C)=mn'+nm' . (4)
In particular,
m= (C,F) , n= (C,E). (5)
This points to the geometric meaning of m and n: just as the degree of a
plane curve is equal to the number of points of its intersection with a line,
so m and n are the two "degrees" of the curve C with respect to the two
systems of rectilinear generators E and F on X.
If we take the embedding XC]p3 into account, a curve acquires a new

geometric invariant: the degree. We know that a family of curves on X can
be simply classified by the invariants m and n. Our object now is to obtain
this classification in terms of the invariants degC and ge'
First 6f all we observe that
degC=(C,H) (6)
where H is a plane section of X. For this purpose we find a plane inter-
-secting C transversally. Then degC is equal to the number of points of
intersection of C and H. Next, if xECnH, then H-pex.e, therefore
(H, C)x= 1, from which it is clear that (C, H) is also equal to the number
of points of intersection of C and H. This proves (6).
Now we mention that
(7)
as is immediately clear from (5) and ·the fact that Hand E, and also
Hand F, are transversal at their point of intersection. Substituting this
expression in (6) and applying (4) we find that
degC=m+n. (8)
Observe that m and n are positive for any irreducible curve C except
when C is a line. For if C does not belong to, say, the first family of
rectilinear generators, then by taking any point x E C and a line E of
the first family passing through x we see that C and E are in general
position and that (C, E) = n;;;. (C, E)x > O.
We proceed to the computation of ge' To apply the formula (1) we
have to know the canonical class of the surface X. Let us turn to this
now. We use the fact that X ~]pl X ]pl. It is easy to solve an even more
general problem: to find the canonical class of a surface X = Yl X Y2 ,
where Y1 and Y2 are smooth projective curves. Denoting by n 1 and n2
the projections n 1 : X - Y 1 , n 2 : X - Y2 we consider arbitrarily one-
dimensional differential forms Wl E .Ql(y1), W2 E .Ql(y2) and associate
with them forms n!(wl) and n!(w2) on X. The form w = n!(wl) A n!(w2)
is two-dimensional and its divisor (w) belongs to the canonical class.
It is this divisor we wish to compute.
Let x E X, X = (Yl' Y2)' Yl E Y1, Y2 E Y2, and let tl and t2 be local
parameters on Y1 and Y2 in a neighbourhood of Yl and Y2' Then an
obvious verification shows that n!(t 1) and n!(t 2) form a system of local
parameters for the point x on X. We represent Wl and W2 in the form
W1 =u 1dt 1, W2=u 2dt 2. Then (Wl)=(U 1) and (W2) = (u 2) in a neighbour-
hood of Yt and Y2' Clearly w = n!(u 1) . n!(u2) dn!(t 1) A dn!(t 2), from
which it follows that in some neighbourhood of x
(w) = (n!(ul)) + (n!(u2)) = n!(wl)) + n!(w2)) .
Since this is true for any point x E X, we see that (co) = n!(COl») + n!(co2»)'
or in other words,
(9)
Now we turn to the case X =]pl X ]pl. We know that - 2y E KIP!'
Y E ]Pl. Therefore the formula (9) in our case gives
. - 2(n!(Yl) + n!(Y2») E Kx.
Since n!(Yl) = E, n!(Y2) = F, we now obtain the final formula
-2E-2FEKx . (10)
To find the genus of the curve C '" mE + nF we have to substitute
this formula in (1) and use (4). So we obtain
gc=(m-1)(n-1) . (11)
Thus, the numbers m and n are uniquely determined, to within a permuta-
tion, by the degree and the genus of C. We see that for a given degree d
there exist d+1 families of curves on X:Mo,M1 , ••• ,Md • The genus of
the family Mj is equal to j (d - j) - d + 1, and the families Mk and Ml
have one and the same genus only if j + I = d, that is, if they are obtained
from each other by the automorphism of ]pI X]p1 that interchanges the
factors. The dimension of the family Mj is (j + 1) (d - j + 1) - 1 or,
expressed in terms of degree and genus: g + 2d.
In his "Lectures on the development of mathematics in the nineteenth
century" Felix Klein gives a classification of curves of degree 3 and 4
Fig. 10 Fig. 11
on a hyperboloid as an example of the application of the idea ofbirational

geometry. Our figures, which illustrate curves with d = 4, are taken from
the same source: In Fig. 10 m=n=2, and in Fig. 11 m= 1, n=3.
3. As another application of formula (1) let us fmd out what negative
values the index of self-intersection of a smooth curve C on a surface
of degree 3 in ]p3 can take. Bya result in Ch. III, § 5.4, in this case K = - E,
where E is a hyperplane section. Therefore (1) takes the form
(C 2 )-degC = 2g-2.
Evidently,(C2 )<0 only if g=O and degC= 1, that is, C is a line on the
surface. In that case (C 2 ) = - 1.
4. The Ring of Classes of Cycles. Our theory of intersection indices

of divisors is a special case of a general theory concerning subvarieties
of arbitrary dimension. The concept of a divisor is replaced here by that
of a I-dimensional cycle. This is the name for the elements of the free
Abelian group generated by the irreducible subvarieties of dimension I.
Two irreducible subvarieties Y1 and Y2 , by defmition, are in general
position if all the irreducible components Z; of the intersection Y1 n Y2
are of one and the same dimension and if
codimZ; = codim Y1 + codim Y2 or Y1 n Y2 = {} .

The basis of the theory in this case is that components Zi are provided
with positive integral multiplicities ni(Y1 , Y 2 ). These multiplicities are
not, in general, lengths of certain rings, as in our theory. They are defmed
as sums in which only the first terms are of this form. The whole theory
turns out to be very complicated and requires a considerably larger
apparatus of commutative and homological algebra. The reader can
fmd an account of it in [29].
The cycle ~n;(Yl' Y2 )Z; is called the product of the subvarieties Y1
and Y2 . By additivity this concept extends to any two cycles in general
position. (Two cycles are, by definition, in general position if each
component of one is in general position with every component of the
other.)
The main property of this mUltiplication is its invariance under a
notion of equivalence, which we shall explain presently. It generalizes
the algebraic equivalence of divisors introduced in Ch. III, § 3.5, and is
defmed in a completely analogous manner. Let T be an arbitrary ir-
reducible smooth variety and Z C X x T a cycle such that Z and X x t for
every point tE T are in general position. The set of cycles Ct=Z·(X x t)
is called an algebraic family. Two cycles C 1 and C2 are called
algebraically equivalent if there exists a family of cycles Ct , t E T, such that
Ctl = C1 , Ctl = C 2 for two points t 1 , t2 E T. The set of classes of cycles

under algebraic equivalence forms a group.
Multiplication of cycles on a projective variety is invariant under
algebraic equivalence. There is a theorem on reduction to general
position according to which for two cycles C 1 and C2 there exist C~ and C~
such that C~ is equivalent to C1 , C~ equivalent to C~,.and C~ and C2 are
in general position. These two results make it possible to define a product
of any two classes of cycles.
We denote by m:r the group of classes (under algebraic equivalence)
of cycles of codimension r on a smooth projective variety X. The group
n
21 = EB 21r ,
r=O
n = dim X ,
is a ring if we define multiplication for individual components as we have

done above, and for arbitrary elements by additivity. This ring is
commutative and associative. By the formula for the dimension of an
intersection [formula (4) of Ch. I, § 6.2J,
that is, 21 is a graded ring. It is easy to show that all points of X, regarded
as zero-dimensional cycles, are equivalent, and that the cycle of x, x E X,
is not equivalent to zero. Therefore the group 21n = 7l . u has the standard
generator u: the class of cycles of x E X. The classes of divisors under
algebraic equivalence form the group 21 1, For n elements (J(1' ..• , (J(n E 211
the product (J(l ... (J(n belongs 21n=71u:
(J(1 •.. (J(n = m.u, mE'll.
The number m coincides with the intersection index «(J(1'.'" (J(n), which
we have defined in § 1.
The ring m: is a very interesting, but not very well explored invariant
of X. The group 210 is isomorphic to 7l- a generator of it is X itself. We
have already stated that m:n ~ 7l. The group 211 is finitely generated: this
is the assertion of Theorem D in Ch. III, § 3.5. Whether the remaining
groups 21r are finitely generated is not known. This is called the basis
problem.
Exercises
1. Determine degvm(lPn), where Vm is the Veronese mapping.
2. Let C be a smooth plane curve of degree r on a smooth surface of degree m in ]p3
Determine (C l ) (generalization of Example 3 in § 1.5).
3. Suppose that on a smooth projective surface of degree m in IP3 the divisor of a form
of degree I consists of a single component with multiplicity 1, which is a smooth curve.
Find its genus.
4. Show that the number of solutions of the system of equations
h(Xb1 ), ••• , x~~); ... ; xg), ... , x!;») = 0,

linear in each of the system of variables x~), ... , x~], is equal to (:En i) ljIT(ni!) if the number
I
of equations is equal to L ni . As always, the number of solutions is understood in the
i=l
sense of the corresponding intersection index.
5. Show that the system of n real equations
!,(X!, ... ,X";Y!, ... ,Y,,)=o (i=I, ... ,n),
homogeneous and of odd degree in each system of variables Xl' ... 'X" and Yl' ... 'Y", has
a real non-zero solution provided that 11# 21.
6. Let X be a smooth curve, D the diagonal in X x X [the set of points of the form
(X, x)]. Show that (D2) = -degK x . Hint: Use the fact that D and X are isomorphic.
7. Show that if a smooth curve C lies on a smooth surface X of degree 4 in IP3 and if
(C 2 ) <0, then (C 2 )= -2.
8. Show that the self-intersection indices of smooth curves on a smooth surface of even
degree in IP3 are always even.
§ 3. BirationaI Isomorphisms of Surfaces

In this section we explain how intersection indices can be applied to
establish some basic properties of birational isomorphisms of surfaces.
We begin by deriving some ofthe simplest properties of a (J-process of an
algebraic surface.
1. (J-Processes of Surfaces. Let X be an algebraic surface, ~ E X a

simple point, x and y local parameters at ~, and (J: Y --+ X a (J-process
centred at this point. According to Theorem 1 of Ch. II, §4, there exists
a neighbourhood U of ~ such that V = (J-l(U) can be described by
equations tOy=tlx in UX]pl, where (to:t l ) are coordinates in ]pl.
Furthermore, in the open set to =1= 0 the (J-process is given by the simple
equations
x=u, y=uv (1)
where v=tdto. At any point I]E(J-l(~) the functions u and v-v(l])
form a system of local parameters. We set L = (J - 1 (~). Evidently, the
local equation of the curve L is u = o.
Let C be an irreducible curve on X passing through ~. Just as in
Theorem 1 ofCh. II, §4, in our case the inverse image (J-l(C) of C consists
of two components: L and a curve C', which can be defined as the closure
of (J-l(C -~) in Y. The curve C' is called the characteristic inverse image
of C. We denote it by (J'(C). Now we regard C as an irreducible divisor
on X. Then
0'* (C) = a'(C) + mL, (2)
where a'(C) occurs with the coefficient 1, because a is an isomorphism
of Y - L onto X -~. Let us find the coefficient min (2). For this purpose
we assume that C has ~ as an r-fold point. This means that iffis a local
equation of C in a neighbourhood of ~, then fE m~, f¢m~+l. Now
a*(C) has a local equation a*(f) in a neighbourhood of any point
11 E a-1(~). We set
(3)
where cp is a form of degree r.
Substituting the transformation formulae (1) in (3) we obtain (a*f)(u, v)
= cp(u, uv) + a*1p. Since 1p E me+ 1, we have 1p = F(x, y), where F is a form
of degree r, with coefficients ill m~. Therefore a*(1p) = (0'* F)(u, vu), and
finally,
(0'* f)(u, v) = ur(cp( 1, v) + u(a* F)( 1, v)) ; (4)
since cp(1, v) is not divisible by u, it follows that in (2) m=r is the multi-
plicity of the singular point ~ on C.
Theorem 1. The inverse image of a prime divisor C on X containing the

centre ~ ofa a-process is given by theformula a*(C) = a'(C) +mL,where
a' (C) is a prime divisor, L = 0'- 1 (~), and m is the multiplicity of ~ on C.
2. Some Intersection Indices. We begin with a general property of

birational regular mappings f: Y ~ X of smooth projective surfaces.
Theorem 2. If D1 and D2 are divisiors on X, then
(1)
If l5 is a divisor on Y whose components are all exceptional curves,

then
(f*(D), l5) = 0 (2)
for any divisor D on X.

We denote by SeX the finite set of points at which the mapping f - 1
is non-regular, and we set T = f- 1 (S) (set-theoretically). Then f de-
termines an isomorphism
Y-T~X-S. (3)
If neither Supp D1 nor Supp D2 intersects S and if D1 and D2 are in
general position, then (1) is obvious by virtue of the isomorphism (3).
Otherwise we use the theorem on shifting the support of a divisor away
from points, (Theorem 1 of Ch. III, § 1.3). Let D'l ~ D1 and D~ ~ D2 be
divisors such that (Supp D'l) n S = (Supp D~) n S = 0 and D'l and D~
are in general position. Then (D l , D2) = (D'l' D~), and by what we have
said above, (Di, D;) = (f*(Dl),f*(D;»). Since f*(D'j) '" f*(D j), the equa-
tion (1) now follows.
(2) is equally obvious if (Supp D) n S = 0. The general case reduces
to this one by very similar arguments.
Now we derive corollaries which refer directly to a a-process. We
use the notation of § 3.1.
Corollary 1.
(4)
We consider a curve C C X with the local equation y. According to
Theorem 1, a* (C) = a' (C) + L, and by (1) in § 3.1 it is clear that a local
equation of a'(C) is v. Since a local equation of L is u, we see that
(a'(C), L) = 1, therefore (4) follows from (2).
Corollary 2. (a' (C), L) = m, where m is the multiplicity of a singular
point ~ on C.
This follows at once from (2), (4) and by (2) in §3.1.
Corollary 3.
(a'(C l ), a'(C 2») = (C l , C2) - m l m2 (5)
where ml and m2are the multiplicities of ~ on C l and C 2, respectively.
By Theorem 1 and 2
(C l , C 2) = (a*(C l ), a*(C 2») = (a'(C l ) + miL, a*(C 2»)
= (a'(C l ), a*(C 2») = (a'(C l ), a'(C 2) + m2L)
= (a'(C l ), a'(C 2 ») + m l m2 ,
from which (5) follows.
3. Elimination of Points of Indeterminacy. We can now derive an

important property of rational mappings of algebraic surfaces.
Theorem 3. If <p : X -+]pn is a rational mapping of a smooth projective
surface, then there exists a sequence of a-processes such that the mapping
1p = <pal" . am is regular.
Proof. We know that <pis non-regular only at finitely many points,

(Theorem 3 ofCh. II, §3.1), and Theorem 2 ofCh. III, §1.4, gives a more
detailed description of this set, which we now recall. Let <p = (fo : ... :f,,),
15 = g.c.d. ((fo), ... ,(f,,»), and Di = (fi) -15. Then the set of points of non-
n Supp D
m
regularity of .
We show that d(<p)-;.O. For this purpose we set A=(AO, ... ,An ),
D;,= (to Ai;) -15. Obviously D;,-;.O,D;,,,,Di. We have to find a A
such that Do and D;, have no common components; thend(<p) = (Do, D;,) -;. 0.
By hypothesis, all the Di do not have common components. Hence for
each irreducible component C C Do there exists an i-;.l such that
VdDi)=O. The condition vdD;,»O means that ~Ai·gilc=O, where the gi
C E C. In view of this, vdD J = °

are local equations of the divisor Di in a neighbourhood of some point
for all A in some non-empty set in An + 1.
Therefore, there exists a A belonging to all the open sets corresponding
to all the irreducible curves C C Go, and for it Do and D;, have no common
n
En
components.
If Xo SUPpDi' then all the SuppD" contain Xo. Therefore d(<p) >
if Supp Di is not empty, that is, if the mapping <p is non-regular. In
°
Xo En
that case we denote by a: X' -+ X the a-process centred at the point
Supp Db and we set <p' = <pa. We show that d(<p') < d(<p), and of
course, Theorem 3 follows from this.
For the divisor D = 'LliCi we define the multiplicity of a point ~ on D
as the number m = 'Lmi1i, where mi are the multiplicities of ~ on the curve
°
C i . Evidently, Theorem 1 then becomes true for every effective divisor,
~nd if D -;. 0, then m -;. 0, where m = indicates that ~ ¢: Supp D.
Similarly we set a' (D) = L lia' (CJ Then a* D = a'D + mL.
We denote by Vi the multiplicities of Xo on the divisors Db and we set
v = min Vi' The mapping <p' is given by the functions.li = a*.t; and
(fi) = (a*.t;) = a'(D i) + (Vi - v) L + v L + a* D),
where the divisors D/=a'(Di)+(vi-v)L,i=O, ... ,n, have no common

components.
We choose an i such that Vi = v, and then, by definition,
From the relation a* D; = a'D; + vL and from Theorem 2 it follows that

((a'D;)2) = ((a* D; - vLf) = ((a* D;f) - v2 = (Dr) - v2 , and therefore
d(q/) = d(cp) - v2 . This proves Theorem 3.
Note. In Theorem 3 there is no need to assume that X is a projective
surface. In the proof this property was used only in referring to the fact
that (C, (j)) = 0 for every curve C C X. However, this proposition was
applied only to curves ofthe form a-l(~), which are projective even when
X is not. It is easy to see that the required property holds for such
curves C.
The simplest example of Theorem 3 is the mapping f: JA? ~]pl in
the definition of a projective line: f(x, y) = (x: y). The mapping f is
non-regular at the point ~ = (0, 0). Substituting (1) of §3.1 we see that
f(x,y)=(l:v) at those points belonging to a-l(~) and to the set to#O,
hence fa is regular there.
4. Decomposition into a-Processes. Now we have everything at our

disposal to prove the main result on birational isomorphisms of sur-
faces.
Theorem 4. Let cp: X ~ Y be a birational isomorphism of smooth
projective surfaces. Then there exists a surface Z and surfaces and mapp-
ings.
a;: X;~X;_l (i= 1, ... , l)
V= 1, ... , m)
such that X 0 = X, Yo = Y, XI = Ym = Z, a; and 7:j are a-processes, and
cpa 1 ... al = 7: 1'" 7: m • In other words, the diagram
Z
I \,
"V \:'
Xl Y1
"II \,tl
X------+)Y
'"
commutes.
Theorem 4 is an obvious consequence of Theorem 3 and the following
proposition.
Theorem 5. Let cp: X ~ Y be a regular mapping of smooth projective
surfaces and a birational isomorphism. Then there exists a sequence of
surfaces and mappings a;: Yi ~ Yi -1 (i = 1, ... , l), such that a; is a a-
process, Yo = Y, Yi = X and
We precede the proof of Theorem 5 with some general remarks on

birational isomorphisms of surfaces.
First of all, for an arbitrary rational mapping qJ: X ~ Y, where X is
a smooth surface and Ya projective variety, we can talk of the image
qJ(C) of a curve C C X. For qJ is regular at all points of C except possibly
a finite set S of points. By qJ( C) we mean the closure of qJ( C - S) in Y.
Then the theorem on the existence of exceptional subvarieties for
regular mappings (Theorem 2 of Ch. II, §4) remains true.
Lemma. If qJ : X ~ Y is a birational isomorphism of smooth projective
surfaces and if qJ - 1 is non-regular at a point y E Y, then there exists a curve
C C X such that qJ(C) = y.
Proof. We consider open sets U C X and V C Y on which qJ establishes
an isomorphism, and we denote by Z the closure of the graph of the
isomorphism qJ: U ~ V in X x Y. The projections onto X and Y de-
termine regular birational isomorphisms p: Z ~ X and q: Z ~ Y. Evi-
dently qJ -1 = P q-1 and since qJ - \ by hypothesis, is non-regular at y,
0
neither is q-1.
We may now apply the theorem on the existence of exceptional
subvarieties (Theorem 2 of Ch. II, §4) to the regular mapping q : Z ~ Y.
This theorem shows that there exists a curve D C Z such that q(D) = y.
We set p(D) = C and verify that C satisfies the condition of the lemma.
Actually, we need only verify that dim C = 1, that is, dim C = dim D.
If this were not the case, then p(C) would be a point x E X, and for all
points ZED we would obtain p(z) = x, q(z) = y, that is, z = (x, y), but
this contradicts the fact that D C X x Y is a curve.
Now we proceed to the proof of Theorem 5. We assume that qJ is
not an isomorphism, that is, qJ -1 is non-regular at a point y E Y. We
consider the a-process a: Y' ~ Y centred at y and we define qJ' : X ~ Y'
so that the diagram
L
(\
Y'
(\
(1)
eX cp IY.5l
X E Z ------~l Y
commutes.
The theorem will be proved if we can show that qJ' is a regular mapping.
For from the commutativity of the diagram (1) it then follows that the
subvariety qJ-1(y) is mapped by qJ' into a- 1(y)=L=1P1. From the fact
that q/ maps X onto the whole of Y' it follows that <p -1(y) is mapped
onto the whole of L. Therefore, not all the components of <p -1 (y) are
mapped into a single point. Hence for y' E L the number of components
of (<p')-1(y') is less than the number of components of <p-1(y).
Consequently, by performing finitely many O"-processes we can achieve
that there are no exceptional subvarieties on X, so that our mapping
becomes an isomorphism.
It remains to prove that <p' is regular. Suppose that this is not so.
Then, by the lemma, 1p = (<p') - 1 maps some curve lying on Y' into a point
x E X. From the commutativity of the diagram (1) it follows that this
curve can only be L, so that 1p(L) = x.
According to Theorem 3 of Ch. II, § 3, there exists a finite set EeL
such that 1p is regular at all points y' E L - E. Since O"(y') = y, it follows
from the commutativity of the diagram (1) that also <p(x) = y.
Let us show that the mapping
(2)
is an isomorphism. It is sufficient to show that it is an epimorphism.

Let dx<pex,x <;;; l c e x.y , where l is a line in the plane e y • Y ' Then it follows
from the commutativity of the diagram (1) that
(3)
for all points y' E L - E. However, this contradicts the simplest properties
of a O"-process. For let C be a smooth curve on Y, y E C, and ey,c =F l, for
example, C = V(cw + [3v), where u and v are local parameters at y. Then
according to (2) of§3.1 O"(O"'(C))=C, and O"'(C) intersects L in a single
point y' which on L has the coordinates (- [3: ()(), O"'(C) is smooth at this
point, and 0": 0"' (C) ~ C is an isomorphism. We can choose ()( and [3 so
that y' <tE, and then already (dy,O")(ey"O",d ct.!.
The facts that (2) is an isomorphism and that <p -1 is non-regular
at y contradict each other. For by applying the theorem on exceptional
subvarieties (Ch. II, §4.2) we can find a curve Z C X, with x E Z, such that
<p(Z) = y. Then ex,z C ex,x (we recall that the tangent space is also
defined when x is a singular point on Z). Since <p(Z) = y, we see that
(dx<p)ex,z = 0, hence that the mapping (2) has a kernel. This contradiction
proves Theorem 5.
5. Notes and Examples. We consider a birational isomorphism
f: X ~ Y of smooth projective surfaces which is also a regular mapping.
We assume that f- 1 is non-regular only at a single point IJ E Y and that
the curve C = f-1(IJ) is irreducible. According toTheorem 5 fis a product
of O"-processes: f = 0"1'" O"m' and since in every O"-process there arises
a curve that is contractible to a point, C is irreducible only if m = 1 and f
itself is a O"-process. Then C coincides with the curve L for which we

proved in §3.1 and §3.2 that
(1)
The converse is also true: if C is a curve on a smooth projective

surface X satisfying the conditions (1), then there exists a regular mapping
f: X ~ Y which is a birational isomorphism, such that Y is smooth,
f( C) = 1] E Y, and that f coincides with a O"-process. Thus, the conditions
(1) are necessary and sufficient for a curve C to be contractible to a point
in the sense indicated above. This result, which is due to Castelnuovo,
will not be proved here; the reader can find a proof in the book [3],
Ch. II.
In conclusion we construct, in accordance with Theorem 4, a de-
composition into O"-processes for a simple birational isomorphism.
This is the birational automorphism f of the projective plane 1P z, the
so-called quadratic transformation, which is given by the formulae:
f(xo : Xl: Xz) = (yo: Y1 : yz)'
(2)
Yo = Xl Xz, Y1 = XoXz, Yz = XOX1 ,
We regard f as a birational isomorphism of two copies: 1P z and ]j52

of the projective plane 1P z, and we denote in one of them the coordinates
by (xo: Xl: Xz), in the other by (yo: Y1: Yz). Clearly fis non-regular at the
three points ¢o = (1: 0,0), ¢1 = (0: 1: 0), ¢2 = (0: 0: 1). According to
Theorem 3 we have to start by carrying out the O"-processes 0"0' 0"1 and
O"z at these points. We arrive at a surface X and a regular mapping
cp:X~1Pz, CP=0"20"10"0' We prove that the mapping 1p=fcp:X~]j5z
is regular. For 1p is regular at a point z if cp(z) =1= ¢i' At the points (E 0"0 1¢o
themappingfO"o is also regular. To verify this itis sufficient to setx = xdxo,
Y = xz/x o, and to substitute (1) of §3.1 in (2). So we see that
f(x, y) = (xy: y: x), Icro(u, v)=(uv: v: 1). (3)
Since 0"1 and 0"2 induce isomorphisms in neighbourhoods of the points

(, we see that 1p is regular at the points z for which cp(z) = ¢o. The situation
is similar for ¢ 1 and ¢ z.
According to Theorem 4 1p is a product of O"-processes: 1p = 1: 1'" 1: m .
Let us clarify what curves C C X can be mapped by 1p into points. Evidently
this can only be either curves M; =O"i- 1(¢i) (i=O, 1,2) or characteristic
inverse images of such curves L C 1P z that are mapped by f into points.
It is easy to see thatf determines an isomorphism between 1P z - Lo - L1
- L2 and ]j5 z - M 0 - M 1 - M z' where Li is the line in 1P2 defined by
the equation Xi = 0, and Mi the line in [>2 with the equation Yi = 0.
Therefore 1p can contract to a point only the curves M~, M'l' M~, L~,
L~, L~ where Li are the characteristic inverses of the curves L; in X.

But from (3) we see that, for example, M~ (given by the local equation
u = 0) is mapped to the whole curve Yo = O. Similarly Mi is mapped onto
M; for i = 1,2. Thus, 1p can shrink into points only the curves Li. Further-
more, 1p-l is non-regular at the points '10=(1:0:0), '11=(0:1:0),
'12 = (0 : 0 : 1), otherwise f- 1 would be regular at one of these points, but
f- 1 is given by the same formulae as f, as is clear from (2). Thus, on the
one hand, in the decomposition 1p = 't 1 ... 'tm there cannot occur more
than three u-processes, and on the other hand, there must occur the
u-processes at the point '10' '11' '12' So we see that
f = 't 2 't 1 'touolUl 1 U21
It is easy to give an idea of the disposition of the curves M~, M'l' M~,
L~, L~, L~,on X. The arrows in Fig. 12 indicate into what points the
curves contract.
Fig. 12
Of course, the quadratic transformation depends on the choice of the

system of coordinates in ]p2 or, what is the same, on the choice of the
points ~o, ~1' ~2' By multiplying out several such transformations we
obtain new birational automorphisms of the plane. Noether has proved
a theorem to the effect that every birational automorphism of the plane
can be represented as a product of quadratic transformations and a
projective transformation. We do not give the very subtle proof of this
theorem, which can be found in [3J, Ch. V. The interesting question to
what extent the representation of a birational automorphism in terms
of quadratic ones is unique has not been investigated so far.
Exercises
1. For every integer 1 (positive, negative or zero) construct a smooth projective
surface X and on it an irreducible curve C such that (C Z ) = I. Hint: Obtain X by blowing
up some points on ]p2.
2. Let X be a smooth projective surface, C I and C z two curves on it. Let x be a
non-singular point on both C I and C z . Let (i: Y -+X be a (i-process at x, C'I and C'z char-
acteristic inverse images of CI and C2 . Show that C~ and C'2 intersect at points Y E IT-I(x)
if and only if C I and C 2 touch at x. Here IT-1(x)n C't n C2 consists of the single point Y
and the order of contract of C~ and C~ at y is one less than the order of contact of CI and
C 2 at x.
3. Suppose that a mapping f: 1P2 ..... IP I is given by the formula
f(x o : Xl: x 2) = (P(x o, Xl' x 2): Q(x o, Xl' x 2)),
where P and Q are forms of degree n. How many IT-processes have to be performed so as to
obtain a surface cp: X ..... 1P 2 for whichfcp is regular?
4. Let X C 1P3 be a smooth quadric and f: X ..... 1P2 the birational isomorphism
consisting in the projection of X from a point X E X. Decompose f into a product of
IT-processes.
5. Letfbe the birational automorphism of 1P2 given in inhomogeneous coordinates
by the formulae x' = X, y' = y + x 2 • Decompose f into a product of IT-processes.
6. Let L C]p2 be a line, X and y two points of it, X ..... 1P2 the product of IT-processes
at X and y, and L' the characteristic inverse image of L. Show that (L')2 = - 1. By Castel-
nuovo's theorem stated in §3.5 there exists a regular mappingf: X ..... Y that is a birational
isomorphism and contracts L' to a point. Construct it in the given case. Hint: Search for
it among the preceding exercises.
7. Letf: X ..... Y be a regular mapping of smooth projective varieties and a birational
isomorphism. Show that for DI , ... , Dn E Div(Y)wehave(f*(D I), ... ,f*(Dn)) = (D I , ... , Dn).
8. Let IT: X ..... Y be a IT-process centred at a point YEY, r=IT-I(Y),DI, ... ,Dn _ 1
E Div(Y). Show that (r, IT*(D I), ... , IT*(D n - t )) = O.
9. In the notation of Exercise 8 find F" for every n> 1.
10. Let X be a smooth projective surface, and IT: X' ..... X a IT-process. Show that the
canonical classes Kx and Kx' of surfaces X and X' are connected by the relation
Kx' = IT* Kx + L, where LeX' is an exceptional curve.
11. Let X be a smooth projective surface, C C X an irreducible curve of which it is
known that its characteristic inverse image relative to finitely many IT-processes is a smooth
curve (it can be shown that such a sequence of IT-processes exists for every curve). Show that
g(C')= (C,C+K) +1-E li(I!-1)

2 2
where C is the normalization of C, and Ii are the multiplicities of all singular points at which
the IT-processes are performed.
12. Show that if C C 1P2 is a curve of degree n with d simple singular points, then
g(C)- (n-1)(n-2) -d
- 2 .
13. Generalize the concept of class (Exercise 20 to Ch. III, §5) to curves with singular
points. Show that under the conditions of Exercise 12 the class ofC is n(n - 1) -d.
Part II. Schemes and Varieties
Chapter V. Schemes
In this chapter we return to the starting point of our entire investigation

- the concept of an algebraic variety, and we attempt to throw light on it
from a more general and invariant point of view. This leads us, on the
one hand, to new concepts and methods, which are exceptionally fruitful
even in the study of quasiprojective varieties with which we have been
concerned previously. On the other hand, we arrive in this way at a
generalization of this concept, which makes the field of applicability of
algebraic geometry far more extensive.
What causes us to look again at the definition of an algebraic variety?
If we recall how affine, projective, and quasiprojective varieties are
defined, we see that in the last analysis they are all defined by systems of
equations. Of course, one and the same variety can be given by various
systems of equations, and it is precisely the wish to abstract from the
accidental choice of the system of equations and of the embedding in an
enveloping space that leads to the concept of isomorphism of varieties.
In this form the system of basic concepts of algebraic geometry reminds
us of the theory of finite extensions of fields at the time when it was
formulated in terms of polynomials: the basic object was the equation
and the idea of independence of the accidentally chosen equation was
formulated by means of the "Tschirnhaus transformation". In the
theory of fields an invariant formulation of the basic concept was
connected with the analysis of a finite extension K/k which, although
(for separable extensions) it can be represented in the form K = k(8),
f(8) = 0, nevertheless reflects the properties of the equation f = 0 that
are invariant under the Tschirnhaus transformation. As another parallel
we can point to the concept of a topological manifold, which in the
works of Poincare was still defined as a subset of a Euclidean space,
until it was invariantly defined as a special case of the more general
concept of a topological space.
The central problem of this and the following chapter are the formula-
tion and investigation of the "abstract" concept of an algebraic variety,
independent of its concrete specification. This concept therefore has in
222 Chapter V. Schemes
algebraic geometry the same function as that of a finite extension in the

theory of fields or of a topological space in topology.
The path on which we are led to this definition is based pn two
observations on the definition of quasiprojective varieties. Firstly, the
basic concepts (for example, that of a regular mapping) are defined for
quasiprojective varieties, starting from their covering by affine open sets.
Secondly, all the properties of an affine variety X are reflected in the ring
k[X], which is invariantly connected with it. These arguments indicate
that the general concept of an algebraic variety must in some sense
reduce to that of an affine variety. In the definition of affine varieties we
have to start out from rings of a certain special type and define a variety
as a geometric object connected with this ring.
The programme we have outlined is not hard to put into effect: in
Ch. I we have investigated in detail how properties of an affine variety
X are reflected in the ring k[X], and this enables us to come to a definition
of the variety X, starting out from some ring which post factum turns
out to be k[X]. However, in this way we may obtain considerably more
than an invariant definition of an algebraic variety. The fact of the
matter is that the coordinate rings of affine varieties are of a very special
kind: they are finitely generated algebras over some field and have no
nilpotent elements (according to Exercises 1 and 2 in Ch. I, §2, all such
rings arise from affine varieties). But since the just developed definition
of an affine variety, starting out from some ring A, satisfies these three
conditions, it makes sense at once to replace in this definition A by a
perfectly arbitrary commutative ring. So we arrive at a wide generaliza-
tion of affine varieties. Since the general definition of an algebraic variety
reduces to that of an affine variety, the concept of an algebraic variety
is also generalized to the same extent. The general concept at which we
arrive in this way is called a scheme.
The concept of a scheme enables us to include an incomparably
wider range of objects than algebraic varieties. We can point to two
reasons why this generalization has turned out to be so exceptionally
useful, both for the "classical" algebraic geometry and for other domains.
Firstly, the rings occurring in the definition of a scheme (analogues to the
rings K[U], where U is an affine open subset of an algebraic variety)
now need not by any means be algebras over a field. For example, they
may be of the type of the ring of integers 7l, the rings of integers in al-
gebraic number fields, or the polynomial rings 7l[T]. The introduction
of these objects makes it possible to apply the theory of schemes in
number theory and provide the best of all the known methods of using
geometrical intuition in number-theoretical problems. Secondly, the rings
occurring in the definition of a scheme may contain nilpotent elements.
For example, the use of these schemes makes it possible to apply in
§ 1. Spectra of Rings 223
algebraic geometry the concepts of differential geometry connected

with infinitely small changes of points or subvarieties Y C X on an
algebraic variety, even when X and Yare quasiprojective varieties. One
should not forget the fact that as a special case of schemes we obtain an in-
variant definition of an algebraic variety, which as we shall see is vastly
more convenient in applications, even when it does not lead to a more
general concept.
In the hope that the reader has already in his possession sufficient
factual material, we abandon the usual style of our book "from the
special to the general". In Ch. V we introduce the general concept of a
scheme and establish its simplest properties. In Ch. VI we define "abstract"
algebraic varieties, which we simply call varieties. By several examples
we show how the concepts and ideas introduced in this chapter
enable us to solve some specific problems which we had encountered
earlier on several occasions in the theory of quasiprojective varieties.
§ 1. Spectra of Rings
1. Defmition of a Spectrum. Let us proceed to the execution of the
programme outlined. We consider a ring A, which we always assume
to be commutative, with an identity; otherwise it is completely arbitrary.
We try to connect with it a certain geometric object, which in case A is
the coordinate ring of an affine variety X must bring us back to X. To
begin with, this object is defined only as a set, but later we endow it with
a number of structures (for example, a topology), which must justify
its claim to be geometrical.
The very first definition requires as a preliminary some explanations.
If we wish to recuperate the affine variety X starting out from the ring
k[ X], then it is most natural to utilize the connection between subvarieties
Y C X and their ideals ay C K[X]. In particular, to a point x E X there
corresponds a maximal ideal M x' and it is easy to verify that the mapping
x~Mx C k[X] establishes a one-to-one correspondence between points
x E X and maximal ideals of the ring k[X]. Therefore it appears natural
to associate with every ring A as a "geometrical object" the set of its
maximal ideals. This set is called the maximal spectrum of A and is
denoted by Specm A. However, in the generality in which we have
considered the problem, the mapping A ~ Specm A has certain defects,
one of which we wish to examine.
Clearly it is to be expected that the association of a ring A with some
set should have the basic properties that link the coordinate ring of an
affine variety with the variety itself. The most important of these properties
is that homomorphisms of rings correspond to regular mappings of a
variety. Is there a natural way of associating with a homomorphism of

rings f:A~B a mapping of SpecmB into SpecmA? More generally,
how can we associate with an ideal be B an ideal a C A? Clearly there
is only one reasonable method, namely to consider the inverse image
f- 1 (b) of this ideal. But alas- the inverse image of a maximal ideal is not
always maximal. For example, if A is a ring without divisors of zero, but
not a field, and iff is an embedding of it in a field K, then the inverse
image of the zero ideal, which is maximal in K, is the zero ideal in A,
which is not maximal.
This complication does not arise if instead of maximal ideals we
consider prime ideals; an elementary verification shows that the inverse
image of a prime ideal under any homomorphism is prime. In the case
when A = k[X] and X is an affine variety, the set of prime ideals of A
has a clear geometrical meaning-it is the set of all irreducible closed
subvarieties of X: points, irreducible curves, surfaces, etc. Finally, for
a very wide class of rings the set of prime ideals is determined by that of
maximal ideals (Exercise 8). All this motivates the following definition:
The set of prime ideals of a ring A is called its spectrum and is denoted
by Spec A. The prime ideals are called the points of the spectrum.
Since we consider only rings with an identity, the ring itself will not
be counted among its prime ideals (so that the factor ring always exists).
Each ring has at least one maximal ideal. This follows easily from
Zorn's lemma (see, for example [37], Vol. I, Ch. III, §8). Therefore the
spectrum of any ring is non-empty.
We have already talked of the geometrical significance of Spec A,
when A is the coordinate ring of an affine variety. Let us look at some
other examples.
Example 1. Spec Z consists of the prime ideals (2), (3), (5), (7), (11) ...
and the zero ideal.
Example 2. Let @x be the local ring of a point x on an irreducible
algebraic curve. Then Spec @x consists of two points: the maximal
ideal and the zero ideal.
Consider a ring homomorphism cp: A ~ B. Henceforth we shall
always be concerned only with such homomorphisms that carry the
identity of one ring into that of the other. As we have remarked above,
for any prime ideal of B its inverse image is a prime ideal of A. By associat-
ing with every prime ideal its inverse image we determine a mapping
acp: Spec B~Spec A,
which is said to be associated with the homomorphism cpo
Example 3. Consider the ring Z[i], i2 = - 1, and let us try to find its
spectrum, by using the embedding cp: Z~Z[i]. This determines the
mapping
aqJ: Spec Z[i] ~ Spec Z .
We denote by wand w' the points of Spec Z and Spec Z [i] that
correspond to the zero ideal. Clearly (aqJ) -1 (w)=w'. Other points in
Spec Z correspond to prime ideals. By definition (aqJ)-l(p) consists of the
prime ideals of Z[i] that divide p. It is standard knowledge that these
ideals are principal and that there are two of them if p == 1 (mod 4) and
one if p = 2 or p == 3 (mod 4). All this can be illustrated as follows (see
Fig. 13):
(2+i) (3-2il
Spec Z[i]
(1+0 (3) (7) (11 )
(2-j) (3+2j)
Spec Z
(2) (3) (5) (7) (11)
Fig. 13
(13) -
w
The reader is recommended to work out the more complicated

example of the spectrum of the ring Z[T], using the embedding
Z~Z[T].
Example 4. We recall that a set SeA that does not contain 0 and is
closed under multiplication is called multiplicative. For every multi-
plicative set we can construct the ring of fractions As consisting of the
pairs (a, s), a E A, s E S, with an identification according to the rule:
(a, s) = (a', s')
if there exists an element s" E S such that

s"(as' - a's) = O.
The operations are defined by the natural rules
(a, s) + (a', s') = (as' + sa', ss') ,
(a, s) . (a', s') = (aa', ss') .
The reader can find a more detailed description of this construction in
[37], Vol. I, Ch. IV, §9. From now on we write a pair (a, s) in the form
a/so In particular, if S is the set A -:p, where :p is a prime ideal, then the
ring As is the same as the local ring Ap of:p.
Suppose S is a nonempty multiplicative set and SE S. The correspond-
ence a~(as, s) determines a homomorphism
cp :A~As,
hence a mapping
acp : Spec As ~ Spec A .
The reader can easily verify that acp is an embedding and that
acp (Spec As) = Us is the set of all prime ideals of A that do not contain
any element of S. The inverse mapping 1p: U s~Spec As has the form
1p(p) = {xis; x E p, S E S}. In particular, iffE A is a non-nilpotent element,
then the system S = {f", n = 1,2, ... } does not contain O. In this case the
ring As is denoted by A f .
2. Properties of the Points of a Spectrum. With every point x E Spec A

we can associate the field of fractions of the quotient ring by the corre-
sponding prime ideal. This field is denoted by k(x). Thus, there is a homo-
morphism
A~k(x) ,
whose kernel is the prime ideal we have denoted by x. The image of

an elementfE A under this homomorphism is denoted by f(x). If A = k[X],
where X is an affine variety and x E X determines a maximal ideal of A,
then k(x) coincides with k, and for f E A the element f(x) defined above
coincides with the value of the function f at the point x. In the general
case every element f E A also determines a "function" on Spec A:
x~f(x) ,
however, with the proviso that its values at distinct points belong,
speaking generally, to distinct sets. For example, for A =Z we can
regard every integer as a "function" whose value at the point (P) belongs
to the field Z/(P), and at the point (0) to the field of rational numbers <Q.
Here we come across one of the most serious difficulties, where the
"classical" geometrical intuition turns out to be inapplicable in our
more general situation. The fact of the matter is that the element f E A
is not always uniquely determined by the corresponding function on
Spec A. For example, the elements to which there corresponds the null
function are those that are contained in all prime ideals. They have a
very simple characterization:
Proposition. An element of A belongs to all prime ideals if and only if it
is nilpotent.
Clearly the proof requires only that an element belonging to all
prime ideals is nilpotent. Letfbe a non-nilpotent element. Then we can
construct the ring A f defined in § 1.1. The image of the points of the
spectrum of this ring under the embedding
SpecAf~SpecA
consists of the prime ideals of A that do not contain any power off, in
particular, f itself. This proves the proposition.
Thus, the inapplicability in the general case of the "functional"
point of view is connected with the presence of nilpotent elements in the
ring. The set of all nilpotent elements of a ring forms an ideal, the so-
called nil-radical of the ring.
With every point x E Spec A there is connected the local ring (l) x:
this is the local ring of the corresponding prime ideal. For example, if
A = Z, then for x = (p) the ring (l) x consists of the rational numbers
whose denominator is prime to p (the p are prime numbers) and (l) x = <Q
for x = (0).
This invariant of a point of a spectrum enables us to carryover to
our general case a number of new geometrical concepts.
For example, the definition of simple points is connected with a
purely algebraic property of their local rings. This suggests the following
definition:
A point x E Spec A is called simple or regular if the local ring (l) x
is Noetherian and regular.
We recall that, in general, Spec A =I- Specm A. Let A = k[X] and
suppose that a point of Spec A corresponds to a non-maximal prime
ideal, that is, to an irreducible subvariety Y C X of positive dimension.
What is the geometrical meaning of regularity of this point of the spec-
trum? As the reader can easily verify, regularity in this case means that
Y is not contained in the subvariety of singular points of X. Let mx be
the maximal ideal of the local ring (l)x of a point x E Spec A. Clearly
and the group mx/m; is a vector space over k(x). If (l)x is a Noetherian
ring (for example, if A is Noetherian), then this space is finite-dimensional.
The vector space
is called the tangent space at the point x of the spectrum.

Example 1. If A is the ring of all integers of some algebraic number
field K (for example, A = Z, K = <Q), then Spec A consists of the maximal
ideals and zero. For x = 0 we have (l) x = K, hence x is a regular point,
and its tangent space is zero-dimensional. But if x = p =I- (0), then (l) x is
known to be a principal ideal ring. Therefore these points are also
regular, and their tangent spaces are one-dimensional.
Example 2. In order to meet with non-regular points we consider the
ring A=Z[mi]=Z+Zmi, where m is an integer greater than 1. The
embedding qJ: A-A', where A' =Z[i], determines the mapping

aqJ : Spec A' _ Spec A . (1)
It is one-to-one if in both rings we restrict ourselves to prime ideals
that are relatively prime to m. It is easy to check that the local rings of
the corresponding ideals coincide. Therefore, a point x E Spec A' can be
non-regular only if the corresponding prime ideal divides m. Of such
ideals there are just as many as they are integral prime divisors of m: if
pi m, then:p = (p, mi) is the corresponding ideal. In this case k(x) = lFp is the
field of p elements and mx/m; = :p/:p2 is two-dimensional over F p.
Therefore, the ideal mx is not principal. Since :p2 C (P), the local ring
~ x is non-regular. Hence all the prime ideals :p = (p, mi), pi m, determine
the singular points of Spec A. The mapping (1) acounts for these singular
points.
To determine the tangent spaces we could naturally go over to
differential forms. The algebraic description of differential forms ex-
plained in Ch. III, § 4.2 enables us to carry them over to arbitrary rings.
We do not need this construction later and do not study it in detail.
3. The Spectral Topology. The topological concepts we have used

in connection with algebraic varieties suggest how to introduce a topology
in the set Spec A. For this purpose we connect with every set E C A the
subset V(E) C Spec A consisting of the prime ideals :p ) E. The relations
V(U E,J=n V(E,J,
a a
V(I) = V(E') U V(EtI) ,
where I is the intersection of the ideals generated by E' and E'; are obvious.
They show that the sets V(E) corresponding to arbitrary subsets E C A
satisfy the axioms for the system of all closed sets of a topological space.
The topology in which V(E) are all the closed subsets of Spec A is
called spectral.
Henceforth, if we talk of Spec A as a topological space, we always
have in mind the spectral topology.
For any homomorphism qJ: A-B and any set E C A we have the
relation
from which it follows that the inverse image of a closed set under aqJ
is closed. This shows that aqJ is a continuous mapping.
As an example we consider the natural homomorphism qJ : A - A/a,
where a is an ideal of A. Clearly, aqJ is a homeomorphism of Spec (A/a)
onto the closed set V(a). Every closed set in Spec A is of the form
V(E) = V(a), where a is the ideal generated by E. Therefore all closed

subsets of Spec A are homeomorphic to spectra of rings.
Let us consider another example. Let SeA be a multiplicative system,
q>:A-+As, Us=aq>(SpecA s ), and 1p:Us-+SpecAs be the mappings and
sets introduced in § 1.1. We equip Us with the topology of a subspace
of Spec A, that is, we take the closed sets to be V(E) n Us. A simple
verification shows that not only aq>, but also 1p is continuous. In other
words, Spec As is homeomorphic to the space Us C Spec A.
Of particular importance is the special case when S = {f', n = 1,2, ... },
where / E A is a non-nilpotent element. Here Us = Spec A - V(f), where
V(f) is that V(E) in which E consists of the single element f
The open sets Spec A - V(f) are called principal and are denoted 'by
D(f). It is easy to check that they form a basis of the spectral topology.
As in the case of affine varieties, the significance of the principal open
sets lies in the fact that they are homeomorphic to the spectra of rings
A J . Using these sets we can prove an important property of spectra:
Proposition. The space Spec A is compact.
What we have to show is that from every covering by open sets we
can extract a finite subcovering. Since the principal open sets form a
U
basis of Spec A, it is sufficient to show this for the covering Spec A = DifJ
The latter condition indicates that nVifa) = V(a) = 0, where a is the
ideal generated by all the elements fa. In other words, there are no
prime ideals containing a, and this means that a = A. But then there
exist elements/a!' ... ,far and gl' ... ,grEA such that
/a,gl + ... +/arg,= 1.
From this, in turn, it follows that ifa,' ... ,fa) = A, that is,
SpecA=DUa)u ... uDUa) .
The spectral topology is very "non-classical", to be precise, very
non-Hausdorff. With such properties of affine varieties we have already
become acquainted in Ch. I, for example: on an irreducible variety any
two non-empty open sets intersect.
This property indicates that the Hausdorff separation axiom is not
satisfied - there exist two distinct points for which any two neighbourhoods
intersect. But owing to the fact that Spec A consists not only of the
maximal, but of all prime ideals, this space is even "less Hausdorff' -non-
closed points exist in it.
ideal peA, then its closure is n

Let us find the closure of a point of Spec A. If our point is a prime
p)E
V(E) = V(p), that is, it consists of the
prime ideals P') P and is homeomorphic to Spec A/p. In particular,
a prime ideal peA is a closed point of Spec A if and only ifit is maximal.
If A has no divisors of zero, then the ideal (0) is prime and is contained
in every prime ideal. Therefore its closure is the whole space - it is an
everywhere dense point.
The existence of non-closed points in a topological space determines
in it a certain hierarchy, which is laid down in the following definitions:
A point x is called a specialization of a point y if x is contained in the
closure of y.
An everywhere dense point is called a generic point of the space.
When does the space Spec A have a generic point? As we have seen
in § 1.2, the intersection of all prime ideals peA consists of all nilpotent
elements of A, that is, it coincides with the nil-radical of the ring. If it is
prime, then it determines a generic point of Spec A. But every prime
ideal must contain all nilpotent elements, that is, the nil-radical. Therefore
Spec A has a generic point if and only if its nil-radical is prime. This
generic point is unique and is determined by the nil-radical.
4. Irreducibility, Dimension. The existence of a generic point is

connected with an important geometrical property of X.
Namely, a topological space X necessarily fails to have a generic
point if it can be represented in the form X = Xl U X 2, where X 1 and X 2
are closed, Xl#- X, X 2 #- X. Such a space is called reducible. For spectra
of rings irreducibility is not only anecessary,but also a sufficient condition
for the presence of a generic point. For it is enough to show that if
Spec A is irreducible, then the nil-radical of A is prime; as we have
shown above, it then follows that Spec A has a generic point. Suppose
that the nil-radical N of A is not prime, and letfg E N,J¢ N, g ¢ N. Then
Spec A = V(f)u V(g) ,

V(f) #- Spec A #- V(g) ,
that is, Spec A is reducible.

Since all the closed subsets of Spec A are also homeomorphic to spectra
of rings, this result carries over to arbitrary closed subsets. Thus, there
exists a one-to-one correspondence between points and irreducible
closed subsets of Spec A. It is determined by assigning to a point its
closure.
The concept of a reducible space leads us at once to a decomposition
into irreducible components. If A is a Noetherian ring, then there exists
a representation
SpecA=X1u ... uX"
where the Xi are irreducible closed subsets and Xi <1 Xj for i #- j, and this
representation is unique. The proof of this fact repeats word for word
the proof of the corresponding proposition for affine varieties, which

was only based on the fact that the ring k[X] is Noetherian.
Example 1. The simplest example of a decomposition of Spec A
into irreducible components is the case of a ring A that is a product of
finitely many rings without divisors of zero:
A=A 1 x ... xA r ·
In this case, as is easy to verify, Spec A is the union of its disjoint irreducible
connected components Spec Ai.
Example 2. To analyse a somewhat less trivial example we take the
group ring Z [a] of a cyclic group of order 2:
A=Z[a]=Z+Za,a 2 = 1.
The nil-radical of this ring is equal to (0), but it is not prime, because
A has divisors of zero: (1 + a)( 1 - a) = O. Therefore
SpecA=X 1 uX 2 , (1)
where Xl = V(1 + a), X 2 = V(1 - a). The homomorphisms qJl' qJ2: A ~Z
with the kernels (1 + a) and (1- a) determine homeomorphisms
aqJl: Spec Z~ V(1 + a),
aqJ2: SpecZ~ V(1-a),
which show that X 1 and X 2 are irreducible and hence that (1) is the
decomposition of Spec A into irreducible components.
Let us find the intersection Xl (\ X 2. Clearly
Xl (\ X 2 = V(1 + a, 1- a) = V(a)
where a is the ideal (1 + a, 1 - a) = (2, 1 - a). Since A/a = Z/(2), we see
that a is a maximal ideal, hence Xl and X2 intersect in a single point
xo = Xl (\ X 2· It is easy to check that if x # x o, for example x E X 1>
X ¢ X 2, then the homomorphism qJ 1 establishes an isomorphism of the
local rings of x and qJ 1 (x). Therefore all the points x # Xo are simple.
But Xo is singular, dim e Xo = 2 and for Y1 = (aqJ 1) - 1 (X o), Yz = (aqJ2) - 1 (X o)
we find that dYl e Yl and dY2 e Y2 are two distinct lines in e Xo : Xo is a
"double point with separated tangents".
It is convenient to represent Spec A by using the mapping
aqJ: Spec A~SpecZ,
where qJ : Z ~ A is the natural embedding (similar to the way in which we

have looked at SpecZ[i] in § 1.1). Then we get the picture of Fig. 14.
-:'
(2) (3) (5) (7) (11) (13)
Fig. 14
- w
Among the purely topological concepts, that is, those that can be
defined by the spectral topology of the ring k[X], there is also the di-
mension of an affine variety X. Of course, the definition we have given in
Ch. 1 as the transcendence degree of the field k(X) uses very specific
properties of the ring k[ X]: that it is an algebra over k, that it is
embedded in a field, and that this field has finite transcendence degree
over k. However, Corollary 3 in Ch. I, §6, puts it into a form that is
applicable to any topological space.
The dimension of a topological space X is the integer n such that in X
there exist chains of distinct irreduc,ible closed sets
XOCX1C",CX n
and there do not exist chains with more terms.
Of course, not every topological space has finite dimension; this is
not even true for spectral rings, even Noetherian ones. Nevertheless, for
a number of important types of ring the dimension of Spec A is finite. In
that case it is called the dimension of A. Without proof we quote the
three main results.
A. If A is a Noetherian local ring, then the dimension of Spec A is
finite and is the same as the dimension of A as it was defined in Ch. II,
§2.1.
B. A finitely generated ring over a ring of finite dimension is itself
of finite dimension.
C. If A is a Noetherian ring, then
dimA[Tl' ... , T,.J=dimA+n.
Proofs can be found in [29J.

Example 3. The ring 7l is of dimension 1. More generally, the ring
of algebraic integers of some algebraic number field is of dimension 1,
because in this ring every prime ideal other than (0) is maximal.
Example 4. To give an example of a ring of greater dimension we
consider the case A =71[T]. In the hope that the reader has already
analysed the structure of the spectrum of this ring in connection with

§ 1.1 we assume the structure of Spec Z[T] to be known. It is very simple:
the maximal ideals are ofthe form (p,f(T)), where p is a prime number
and fEZ[T] is a polynomial which on reduction modulo p becomes
irreducible; the non-maximal prime ideals other than (0) are principal
and are of the form (P) or (f(T)), wherefis a primitive polynomial. From
this it follows that the chains of prime ideals of maximal length are:
(p,f(T)) ) (g(T))) (0)

or
(p,f(T)) ) (P) ) (0) .
Thus, dim Z[T] = 2 in accordance with Proposition C.
Exercises
1. Let N be the nil-radical of a ring A. Show that the natural embedding
aq>: Spec A/N --Spec A is a homeomorphism.
2. Show that an elementfE A is a divisor of zero if and only if there exists a decomposi-
tion Spec A = X u X' into closed subsets X, X' # Spec A such that f(x) = 0 for all points
XEX.
3. Let q> : A ..... B be an embedding of rings and let B be integral over A. Show that
aq> is an epimorphism.
4. Let q>: A ..... B be a homomorphism of rings. Does the mapping aq> always carry
closed points into closed ones? Is this true under the conditions of Exercise 3?
5. Show that aq>(V(E)) = V(q>-I(E)). The bar denotes closure.
6. Let X, and X 2 be closed subsets ·of Spec A and let u" U2 E A be such that
U, +U2= 1, U, U2=0, U;(X) =0 for all points XEXi, i= 1,2. Show that then A=A, x A 2, with
Xi = aq>i(Spec Ai), where q>i: A ..... AI is the natural homomorphism.
7. Let Spec A = X, u X 2 be a decomposition into closed disjoint sets. Show that
then A = A, X A 2, Xi= aq>i (Spec AJ Hint: Representing Xi in the form V(Ei) fmd elements
Vi, i = 1,2, such that V, + V2 = 1, Vi(X) = 0 for all x E Xi. Using the proposition in § 1.2
construct functions U , and U2 satisfying the conditions of Exercise 6.
8. Show that if A is a ring of finite type over an algebraically closed field, then the
proposition of § 1.2 remains valid if in its statement prime ideals are replaced by maximal
ideals. Hint: Use Hilbert's Nullstellensatz.
Deduce that closed points are everywhere dense in every closed subset of Spec A.
9. Let A = Z[T]/(F(T)), where F(T) E Z[T], p is a prime number, F(O):;; O(P) and
pEA is the maximal ideal of A generated by p and the image of T. Show that the point
x E Spec A corresponding to p is singular if and only if F(O):;; 0(P2), F'(O):;; O(P). Hint:
Consider the homomorphism M/M2 ..... p/p2, where M=(p,F)EZ[T].
to. Show that in SpecZ [T" ... , TJ every closed subset each of whose components is
of codimension 1 (that is, of dimension equal to dim SpecZ[T" ... , Tn] -1) is of the form
V(F), where FEZ [T ... , Tn].
"
11. Prove the following universal property of the ring of fractions As relative to a
multiplicative system SeA (Example 4 of § 1.1): iff: A ..... B is a homomorphism for which
all thefts), S E S, are invertible in B, then there exists a homomorphism g: As ..... B for which
f=gh, where h is the natural homomorphism A ..... A s .
§2. Sheaves
1. Presheaves. The concept of the spectrum of a ring is only one of two
elements from which the definition of a scheme is made up. The second
element is the concept of a sheaf. In the preceding section we have used
the fact that an affine variety is given by the ring of regular functions on
it, and starting out from an arbitrary ring we have arrived at the corre-
sponding geometrical concept-the spectrum. In the definition of the
general concept of a scheme we also take as our basis the regular functions
of a variety. But there may be too few of them if we consider functions
that are regular on the whole variety. It is therefore natural to consider
for any open set U C X the ring of regular functions on it. In this way
we obtain not one ring, but a system of rings, between which, as we shall
see, there are various connections. We base our definition of a scheme
on a similar object. However, to begin with we must analyse some de-
finitions and the simplest facts relating to this kind of object.
Definition. Let X be a topological space and suppose that a certain
set $'(U) is associated with every open set U of it, and that for any two
sets U C V there is a mapping
eb: $'(V)--+$'(U).
This system of sets and mappings is called a presheaf if the following
conditions are satisfied:
1) if U is empty, then the set $'(U) consists of a single element;
2) eg is the identity mapping;
3) for any open sets U eVe W we have
elf = ebetf . (1)
Sometimes such a presheaf is simply denoted by the letter $'. If it is
important to emphasize that the mapping eb refers precisely to a presheaf
$', then it is denoted by eb,F.
If all the sets $'(U) are groups, modules over a ring A, or rings, and
if the mappings eb are homomorphisms of these structures, then we speak
of a presheaf of groups, modules over A, or rings.
Evidently a presheaf $' does not depend on a choice of the element
$'(0) (more accurately, for distinct choices we obtain isomorphic
presheaves relative to a concept of isomorphism that the reader can
easily supply). Therefore, to specify a presheaf it is sufficient to indicate
the sets $'(U) for non-empty sets U. If $' is a presheaf of groups, then
$'(0) is the group consisting of one element only.
If $' is a presheaf on X and U C X is an open set, then the assignation
V --+$'(V) for all open sets V C U obviously determines a presheaf on U.
It is called the restriction of $' and is denoted by $'Iu.
§2. Sheaves 235
Examples. 1. Let M be a set, let :F(U) consist of all functions on U

with values in M, and for U C V let llb be the restriction of a function
given on V to U. The relation (1) is obvious. :F is called the presheaf of
all functions on X in M.
In order to carryover the intuitive picture of this example to the case
of an arbitrary presheaf, the llb are called restriction mappings.
Example 1 can be varied.
2. Let M be a topological space, let :F(U) consist of the continuous
functions on U with values in M, and let the llb be as in Example 1.
Then :F is called a presheaf of continuous functions.
3. X is a differentiable manifold, :F(U) the set of differentiable
functions on U (with real values). The mappings llb are again as in
Example 1.
4. X is an irreducible quasiprojective variety in which a topology is
defined by the fact that closed subsets are algebraic subvarieties (so
that the "topological" terminology of Ch. I turns into the usual topo-
logical concepts). For an open set U C X, let :F(U) be the set of all
rational functions on X that are regular at all points of U. The mappings
eb are as in Example 1. Then :F is a presheaf of rings, which is called the
presheaf of regular functions.
2. The Structure Presheaf. Now we go over to the construction of the

presheaf that will play the decisive role in what follows. It is defined on
the topological space X = Spec A. The presheaf we are about to define
is called the structure presheaf on Spec A and is denoted by (9. To make
the logic of our definition clearer we give it first in a more special form.
Let us assume, to begin with, that the ring A has no divisors of zero,
and let K be its field of fractions. Now we can copy faithfully the Example
4 above. For an open set U C Spec A we denote by (9(U) the set of those
elements UE K which for every point XE U have a representation u=a/b,
a, b E A, b(x) =F 0, (that is, b is not contained in the prime ideal x).
Clearly (9(U) is a ring. Since all the rings (9(U) are contained in K,
we can compare them as subsets of a single set. Evidently, if U C V,
then (9(V) C (9(U). We denote by llb the embedding of CD(V) in CD(U).
A trivial verification shows that we obtain a presheaf of rings.
Before completing the discussion ofthis case, let us compute CD(SpecA).
Our arguments repeat the proof of Theorem 4 in Ch. I, §3. The condition
u E CD(Spec A) means that for every point x E Spec A there exist elements
ax and bx E A such that
(1)
Consider the ideal a generated by all the elements bx , x E Spec A. By (1)

it is not contained in any prime ideal of A, so that a = A. Thus, there
exist points Xl' ... , Xr and elements C I , ... , Cr E A such that

C I bx, + ... + crb Xr = 1 .
Multiplying (1) for X = Xi by cib x ; and adding up we find that
Thus, l!J(Spec A) = A.
Now we go over to the case of an arbitrary ring A. The preceding
arguments suggest that it would be natural to set l!J(Spec A) = A. But
there also exist other open sets U that have natural candidates for the
ring l!J(U), namely the principal open sets D(j),fEA. For we have seen
in § 1.3 that D(j) is homeomorphic to Spec A I' and it is equally plausible
to set
l!J(D(j)) = AI.
Thus, so far we have defined the presheaf l!J(U) on principal open sets
D(j). Before defining it on all open sets we introduce the homomorphisms
eb, of course, only for principal open sets V and U.
To begin with, let us find out when D(j) C D(g). This is equivalent
to the fact that V(j) J V(g), that is, every prime ideal containing g also
I
contains f In other words, the image of f in the ring AI(g) is contained
in every prime ideal of this ring. In § 1.2 we have seen that this is equivalent
to I being nilpotent, that is, to f" E (g) for some n > O. Thus, D(j) C D(g)
if and only if for some n > 0 and U E A
f" = gu. (2)
In this case we can construct the homomorphism e~:~) of Ag into
A I by setting
An obvious verification shows that this mapping does not depend

on the representation of an element t E Ag in the form t = algi and that
it is a homomorphism. (This homomorphism can be written down
in a more invariant form, by making use of the universal ring of fractions
As, see Exercise 11 to § 1. In our case g and its powers are invertible in
A lowing to (2), from which the existence of the homomorphism e~:~)
follows.)
Before we formulate the final definition we return for a moment to the
case already considered when A has no divisors of zero. Then we can
indicate a method of computing l!J(U) for every open set U in terms
of l!J(V), where V are the various principal open sets. Namely, if {D(f)}
are all possible principal open sets contained in U, then, as is very easy
to check,
l!J(U) = n l!J(D(j)).
§2. Sheaves 237
In the general case one might wish to take this equation as definition,
but this is impossible, because the (9(D(f)) are not contained entirely in
one common set. But they are connected among each other by the
homomorphisms QZ!~i if D(g) C D(f). In this situation a natural generaliza-
tion of the intersection is the projective limit of sets. Let us recall its defini-
tion. Let I be a partially ordered set, {E"" ex E I} a system of sets, and ff for
any ex, /3 E I, ex,;;;, /3, a mapping of Ep into E", satisfying the following
conditions:
1) .t: is the identity map of E"" and 2) for ex,;;;, /3,;;;, y we havefl = ftfJ.
n
Consider the subset of the product E", of the sets E", consisting of those
"eI
elements x={x",;x",EE",} for which x,,=ftxp for all ex,;;;,/3. This subset
is called the projective limit of the system of sets E", relative to the system of
-- --
homomorphisms ft and is denoted by lim E",. The mappings x ~ x"'
E lim E", of the projective limit are said to be natural.
If the E", are rings, modules, or groups, and if the ff are homomorph-
isms of these structures, then lim E", is a structure of the same type.
The reader can find a more detailed description of this construction
in [12], Ch. VIII, §3. One has to keep in mind here that the condition
on the set I to be directed is not essential for the definition of the projective
limit.
Now we are prepared for the fmal definition:
--
(9(U) = lim (9(D(f))
where the projective limit is taken over all D(f) C U relative to the system
of homomorphisms QZ!~1 for D(g) C D(f) we have constructed above.
By definition, (9(U) consists offamilies{u",},u"EAf~' whereh are all
those elements for which Difa) C U, and
(3)
For U C V every family {v,,} E (9(V) consisting of V",E Af~' Difa) C V,

determines a subfamily {vp} consisting of the vp with index /3 such that
D(fp) cU. Evidently {vp} E (9(U). We set
Q:;({ v",}) = {v p} .
A trivial verification shows that (9(U) and Q:; determine a presheaf (9,
the so-called structure presheaf, on Spec A.
If U = Spec A, then D(1) = U, so that 1 is one of the f", say fo. The
mapping
{u,,}~uo
determines, as is easy to check, an isomorphism (9(Spec A) ~ A.

In particular, if U = {u a; D(fJ C U} E (9(U), then by definition

eE(f)u= {up;D(fp) C D(fn. By what was said above, the assignation
{up, D(fp) C D(f)} """"U a if f = fa determines an isomorphism between
(9(D(fa)) and Afo by which
(4)
This formula allows us to recover Ua from the element U E (9(U) deter-
mined by it. '
3. Sheaves. We assume that the topological space X is the union

of open sets U(I.' Every function f on X is uniquely determined by its
restrictions to the sets U a' and if on each of the U(I. a function h is given
for which the restrictions offa andfp agree on U a n Up, then there exists
a function f on X such that every fa is its restriction to U a' Continuous
functions, differentiable functions on a differentiable manifold, and regu-
lar functions on a quasiprojective algebraic variety have the same pro-
perty, which expresses the local character of the concept of being a
continuous, differentiable, regular function. It can be rephrased for any
pre sheaf and singles out an exceptionally important class of presheaves.
Defmition.A presheaf!!J' on a topological space is called a sheaf if for
every open set U C X and any open covering of it U = UUa the following
conditions are satisfied:
1) if egosl = egs2 for SI, S2 E !!J'(U) and all U a, then SI = S2;
2) if saE!!J'(Ua) are such that eg:f"lUps(I.=eg:f"lUpSp, then there
exists an S E .~(U) for which S(I. = egos for all Ua .
We have already given a number of examples of sheaves before
defining this concept. We now indicate the simplest example ofa presheaf
that is not a sheaf. Let X be a topological space, M a set, !!J'(U) = M for
all U C X, and e~ the identity mapping. Evidently .'F is a pre sheaf.
Suppose that in X there exist disconnected open sets and that U = U I U U 2'
U I n U 2 = 0 is a representation of one of them as the union of disjoint
open sets. Let ml and m 2 be distinct elements of M and SI = ml E .'F( U I)'
s2=m 2 E!!J'(U 2). Evidently the condition eg:f"lU2sl=gg~f"lU2S2 holds
automatically, however, there does not exist an S E !!J'(U) such that
gg,S=SI' eg2s=S2, because m l =!=m 2 .
Theorem 1. The structure presheaf on Spec A is a sheq[.

To begin with, we check the conditions 1) and 2) in the definition of
a sheaf for the case when U and the U(I. are principal open sets.
First of all,we note that both conditions need only be verified for the
case U = Spec A. For if U = D(j), U(I. = D(fa), then, as the reader can
easily verify, the conditions 1) and 2) hold for U and U a if they hold for
Spec Af and the sets [j a = Deja), where 1(1. is the image of h under the
§2. Sheaves 239
natural homomorphism A--+Af . Now we proceed to the verification of

1) and 2) for U,,=D{h),UU,,=SpecA.
1. Since (!) is a presheaf of groups, it is sufficient to show that if
UE (!)(Spec A) = A and {!~ecAU = 0 for all UIX' then U= O.
The condition (!~~ecAU = 0 means that
.t:"'U=O (1)
for all IX and certain nIX ~ O. Since D(fJ = D(J:"'), we have UDU;"') = Spec A.
We have already seen that this leads to an identity
.t:,'gl + ... +.t:;gr = 1
for suitable gl, ... ,grEA. Multiplying (1) for 1X=1X1, ... ,lXr by gl, · .. ,gr
and adding up we find that U = O.
2. Since the space Spec A is compact, we may restrict ourselves to
the case of a finite covering. For the reader can easily verify that if the
assertion is true for subcoverings, then it is also true for the whole
covering.
Let Spec A = D(fl) U ... u D(f,.) and Ui E A fi' Ui = Vi!Jr (a single n may
be chosen because the covering is finite). First of all, we observe that
D(f) n D(g) = D(fg) (the verification is trivial). By definition
D(f,) _ vJj
{!D(f;fj)u i - (fJj)n'
and by hypothesis,
(fJjr(vJj - vji') = 0 .
Setting vJ".r= Wj' m+ n= k, we find that
Ui = wJft, wff= Wfl· (2)
As in the verification of 1), we see that
"L./;kgi = 1 .
We set U= "L.Wjgj. By hypothesis,
/;"u = L wjgJ"l= L wigjI= Wi'

j j
(2)
'" (!D(fi)
Therel0re Spec A
U -- WilJi - Ui .
11'1<_
The verification of 1) and 2) for arbitrary open sets is a formal con-

sequence of what we have already proved. Our situation can be described
in general terms as follows. On the topological space X a basis 1/ = {V;}
of open sets is given that is closed under intersections. We assume that
thepresheaf ff of groups on X satisfies the conditions: a)ff(U) = limff(V,,~
+-
where the limit is taken over all E"Y, r-: v:

c U, relative to the homo-
morphisms e~;, and b) the e~~ coincide with the natural homomorphisms
of the projective limit. Both these properties are satisfied for the presheaf
(rJ-the first by definition, and the second by (4) in §2.2. We show that under
these conditions :11' is a sheaf if 1) and 2) hold for the sets V. E Y.
1. Let U = U U~= U
~,A' ~,A E Y. If eg"u =0 for all U~, then
~ ~,A
e~",,,u=O. Introducing new indices ((,,1)=Y, we obtain U=UV;"

e~y u = 0 for all v;..
To show that u = 0 we need only verify by condition b)
that e~ ~u = 0 for all v: c
U. This follows at once from an analysis of
the homomorphisms corresponding to the sets
U'\)
Indeed,
nV~
K~n~K~
nU U--K~n~
nU nV'
U--K~n~K~
nU U-- 0
v:
for all v;., hence e~ aU = 0 because = U(V: (\ v;.), and 1) holds by hypo-
thesis for the sets Va.
2. Let u~ E :1I'(U ~), eg~: nU", U~l = eg~: nU", U~2' U ~ = U ~,A' Setting
).
Vp = e~~,).u~ and Y = ((, A) we verify that

(3)
This follows from the analysis of the homomorphisms e corresponding
to the sets
U U
U ~1'-0 U ~2 ~
v;. 1 ~ U~lG U~2 ~ VY2
v;.1 (\ v;.2
Yl = (( l' ,.1,1)' Y2 = ((2' ,12)
The left-hand side of (3) is equal to
n U", U n U", nU", O~",
= t:::VYlnVY2IM.D';lnUl;2 U
~V-YlnVY2 ~1 ~1·
Evidently it is equal to the right-hand side. By (3), for every E i/, v:

v: C U the elements e~~nVyVy satisfy a similar relation, hence by hypo-
thesis there exist elements vaE:1I'(V,) for which e~:nVyVa=e~~nVyvl"
An obvious verification shows that these elements determine an element
u of the projective limit ~ .~(V:). For it e~~u = Va' Hence for u~ = eg"u
§2. Sheaves 241
we have Q~!u~ = Q~~u~, for all V. C U~, V. E "1', and so u~ = u~. This
completes the proof of the theorem.
4. The Stalks of a Sheaf. We return to the analysis of our concepts

of a sheaf and a presheaf. We consider, to begin with, a presheaf for which
all the sets ff(U) are subsets of a single set and the mappings Q~ are
embeddings of ff(V) in ff(U); this is so, for example, for the sheaf ~ on
the irreducible space Spec A. Then we may consider the union ffx of
sets ff(U) for all open sets U containing the given point x. For the sheaf
of continuous functions ffx these are the germs of the functions con-
tinuous on some neighbourhood of x, that is, the result of identifying
functions that coincide in some such neighbourhood. For the sheaf of
regular functions on an irreducible quasiprojective variety ffx this is
the local ring of x.
In the general case not all the sets ff(U) are contained in a single
one, but they are connected by the homomorphisms Q~, and this enables
us to replace the union by the inductive limit. The definition, similar to
that of the projective limit, can be read in the book [12J, Ch. VIII, §4.
Defmition. The stalkffx of a presheaf ff at a point x E X is the inductive
limit of the sets ff(U) for all U containing x relative to the system of
mappings Q~ for U C V.
By definition, an element of ffx is given by an element of any of the
ff(U), x E U, but here two elements u E ff(U) and v E ff(V), x E U,
X E V, are identified if there exists aWe Un V, x E W, such that
Q~u = Q:;'v.
Applying this definition to the case of the structure sheaf ~ on
Spec A we see that the stalk ~ x is the same as the local ring of the prime
ideal x E Spec A.
In the general case, for every open set U containing x there is defined
the natural homomorphism
(R : ff(U)-+ffx·
If ff is a sheaf and if Q~Ul = Q~U2 for two elements u 1 , U2 E ff(U) and all
points x E U, then U1 = U2. For by definition this means that every point
x E U has a neighbourhood W, We U, such that Q~Ul = Q~U2. By
definition of a sheaf it then follows that Ul = U2.
Thus, for a sheaf ff the elements of ff(U) can be given by families
{u x ; Ux E ffx' x E U}. Of course, as a result we do not obtain all families
of this form. The following condition is clearly necessary:
For every point x E U there exists a neighbourhood We U and an
element WE ff(W) such that uy = Q': W for all points YEW.
The reader can easily verify that, conversely, any family satisfying
this condition corresponds to some element u E ff(U).
This is true, of course, only when ff is a sheaf. But if ff is an arbitrary

presheaf, we may nevertheless consider the set ff'(U) of all families
{u x } satisfying the condition above. For U C V the mapping
Qb: {v x ; VxE ffx' x E V}~{Vy; Vy E ffy,y E U}
turns ff'(U) into a pre sheaf. It is easy to see that, in fact, we obtain in
this way a sheaf. It is called the sheaf associated with the presheaf ff. It is
the sheaf "closest" to ff. For example, if ff is a presheaf for which
ff(U) = M for all U (so that it consists of the constant functions on U
with values in M), then ff'(U) consists of all functions on U that are
constant on each connected component of U.
Exercises
1. Let X be a discrete topological space;' F(U) the collection of those mappings
f: U .... M for whichf(U) is finite, and for U C V let (!t be the restriction. Is :F a pre sheaf?
Is it a sheaf?
2. Let X be a smooth quasiprojective variety in which a topology is introduced as in
Example 4 of §2.1. For an open set U C X we set :F(U) = QP[U], and for U C V we define
(!t as the restriction of differential forms. Is :F a presheaf? Is it a sheaf?
3. Let A be a ring a C A an ideal. For every non-nilpotent f E A we denote by a j the
ideal generated in A j by the images of the elements of a under the homomorphism A .... A j '
By analogy with what we have done in §2.2, construct the presheaf i1" for which .'F'(U) = aj
if U = D(!), and show that:F is a sheaf. A simpler variant: examine the case when A has no
divisors of zero.
4. Let X be a topological space, M an Abelian group, :F(U) the factor group of all
locally constant functions on U with values in M by the constant functions, and let (!t
be defined by means of restrictions. Show that i1" is a presheaf and determine the sheaf
:F' associated with it.
5. Show that the structure sheaf (1J on Spec A can be defined as follows. For the elements
U E (1J (U) we take families of elements {u x ; U x E (1J xo X E U} «(1J x is the local ring of the prime
ideal x) that satisfy the condition: for any point Y E U there exists a principal neighbourhood
Y E D(f) C U and an element U E A j such that all the U x for x E D(f) are images of u under the
natural homomorphisms A j .... (1Jx. If UCV, then by taking in the family v={Vx;XEV}
those Vx for which x E U we obtain (It v.
6. Let A be a one-dimensional local ring and'; E Spec A a generic point. Show that
~ is an open set and find (1J(~).
7. Let A be the local ring of the origin of coordinates in /A2 . Find (1J(U), where
U = Spec A - x and x is a closed point.
§3. Schemes
1. Definition of a Scheme
Defmition. A ringed space is a pair (X, @) consisting of a topological
space X and a sheaf of rings @.
A morphism of ringed spaces cp: (X, @x)~(Y, my) is a collection of a
continuous mapping cp: X ~ Y and homomorphisms
1fJu: @y(U)~ @X(cp-1(U))
§3. Schemes 243
for every open set U C Y for which cp -l(U) is non-empty. It is required

that the diagram
is commutative for any U and V, U C V

Sometimes the sheaf (!) is denoted by (!)x. It is called the structure
sheaf of the ringed space.
Example 1. Any topological space X is a ringed space if we take for
the sheaf of continuous functions. Any continuous mapping cp : X -+ Y
(!)x
defines a morphism if we set lPu(f) = cp*(f) for fE (!)y(U).
Example 2. Any differentiable manifold is a ringed space if we take
for (!) x the sheaf of differentiable functions. Just as above, any differentiable
mapping determines a morphism.
Example 3. Any ring A determines the ringed space (Spec A, (!))

where (!) is the structure sheaf. Henceforth we denote this ringed space
by Spec A. Let us show that a homomorphism A: A -+ B determines a
morphism cp: Spec B -+ Spec A. We set cp = a A. For U = D(f) C Spec A
we have cp-1(U)=D(A(f)). The mapping a/f'-+A(a)/A{nn determines a
homomorphism lPu of the ring A J = (!)SpecA (U) into the ring
BA(f) = (!)SpecB(cp-1(U)). The reader can easily verify that these homo-
morphisms extend to homomorphisms lP: (!)SpecA(U)-+ (!)SpecB(CP -1 (U))
for every open set U C Spec A and define a morphism cp of ringed spaces.
In what follows we frequently denote a ringed space (X, (!)x) by a single
letter X and a morphism X -+ Y given by mappings cp and lPu by a single
letter cp.
A simple check shows that by combining two morphisms cp : X -+ Y
and cp': Y -+ Z (cp as well as lPu) we obtain a morphism cp' cp : X -+ Z.
A morphism having an inverse is called an isomorphism.
If (X, (!) x) is a ringed space and U C X an open set, then by restricting
the sheaf (!)x to U we obtain the ringed space (U, (!)xlu). In this sense we
shall henceforth often regard an open set U C X as a ringed space.
As regards the examples we have given we make two remarks.
1. Whereas in Examples 1 and 2 the morphism is uniquely deter-
mined by the mapping cp: X -+ Y because the corresponding homo-
morphisms lPu are given by inverse images of functions, in Example 3
this is not so. For instance, if the ring A has a non-zero nil-radical N,
B = A/N, and A: A -+ B is the natural projection, then Spec A = Spec B
and cp = a A is the identity mapping; however, even for U = Spec A we see
that 1fJu = A. is not an isomorphism. Thus, a morphism of ringed spaces

does not reduce to a mapping of the corresponding topological spaces.
To emphasize this fact the word "mapping" is here replaced by the
term "morphism".
2. The concept of a ringed space gives a convenient principle of
classifying geometric objects. For example, let us take differentiable
manifolds. They can be defined as ringed vector spaces, namely those in
which every point has a neighbourhood U such that the ringed space
(U, (Olu) is isomorphic to (D, iff), where D is a domain in n-dimensional
Euclidean space and iff is the sheaf of differentiable functions on it.
Precisely this definition is contained, for example, in the book [26],
the only difference being that the terminology of sheaves is not used.
The general idea of this method of determining geometric objects
is the following: we impose restrictions on the local structure of the ringed
space by requiring each point to have a neighbourhood isomorphic, as a
ringed space, to one of the ringed spaces of a previously fixed class.
The last remark leads us to the basic definition:
A scheme is a ringed space (X, (0) in which every point has a neigh-
bourhood U such that the ringed space (U, (01 u) is isomorphic to Spec A,
where A is a ring*.
A morphism of schemes cp: X -+ Y is dermed as a morphism of corres-
ponding ringed spaces, satisfying the following condition: if U c Y is
open, x E f- 1 (U), UE (Oy(U) and u(cp(x))=O, then CPu(u)(x)=O.
If X is a scheme and A a ring, then the morphism X -+ Spec A de-
termines a homomorphism A-+(ox(U) for every open set U C X, that is,
it turns (Ox into a sheaf of algebras over A. It is not hard to show that,
conversely, if (Ox is a sheaf of algebras over A, then it specifies the canonical
morphism X -'+ Spec A. A scheme X for which a morphism X -+ Spec A
is given is called a scheme over A. A morphism of schemes over A is
determined by the condition that the diagram should be commutative,
x~ • /Y
Spec A
and this is equivalent to the fact that all the 1fJu are algebra homomorph-
isms over A.
Since every ring is an algebra over the ring of integers 7L, every scheme
is a scheme over 7L. In this sense the concept of a scheme over A generalizes
that of a scheme.
* If (X, lP) is scheme, U C X open, XE U and UE lP(U), we say that v(x)=O if, for a neighbour-
hood U' C U of x, for which there is an isomorphism l(! :(U', lPl u )-' Spec A, pg.(v) and x are
mapped by l(! to such clements v' E A' and x' E Spec A that v'(x')=O. Obviously this does not
depend on the choice of U'.
§3. Schemes 245
Here are two very simple examples of schemes.
Example 4. Example 3 of a ringed space shows that Spec A is a scheme

for every ring A. Such schemes are called affine. There is a one-to-one
correspondence between ring homomorphisms A.: A - Band morph-
isms A.: Spec B-Spec A.
Example 5. Let us find out how the notion of a quasiprojective variety

fits into the language of schemes. We begin with the case of an affine
variety X over an algebraically closed field k. The scheme Spec k[X]
defined in Example 4 does not coincide with X even as a set: Spec k[X]
consists of all prime ideals of the ring k[X], which in tum corresponds
to all irreducible subvarieties of X, and not only to its points. Neverthe-
less, the variety X and the scheme Spec k[X] are very naturally connected
with each other, if only because regular mappings of affine varieties
X - Y and morphisms of schemes Spec k[X]-Speck[Y] are one and
the same thing: both correspond to algebra homomorphisms k[Y]-
k[X]. Thus, we have here an isomorphism of categories .
. Now let us consider an arbitrary quasiprojective variety X over a
field k and associate with it in a similar fashion a scheme X over k. For
the set X we take the collection of all irreducible subvarieties of X
(including X). Let U C X be an open subset and Uthe set of its irreducible
subvarieties. By associatin~ with ~ subvariety Z_C U_ its closure Z C X
we define an embedding of U into X. The subsets U C X define a topology
in X. Finally, we define the sheaf (!)x by the condition (!)x(U) = k[U]
with the natural restriction mapping. The reader is invited to verify
that in this way we tum X into a scheme over k.
A regular mapping I: X - Y determines a mapping of sets j: X - y-
in which an irreducible variety Z C X corresponds to j(Z), the closure of
I(Z) in Y. Finally, for U C Y we define the homomorphismju: (!)y(U)-
(!)x(1-1(U)) as the homomorphism f*: k[U]_k[f-l(U)]. The reader
can easily verify that j is a morphism of X in Y, and that the assignation
I-j defines a one-to-one correspondence between regular mappings
X - Y and morphisms X - y- as schemes over k. Again we have an iso-
morphism of categories.
In what follows we frequently do not distinguish between a quasi-
projective variety and the scheme corresponding to it.
In conclusion we make a few obvious remarks concerning the
definition of a scheme.
The structure sheaf (!) of a scheme X has one important property:
its stalk (!)x over any point x E X is a local ring. For the stalk :fix of an
arbitrary sheaf :fi on a space X does not change if we replace X by an
open subset U containing x. For the structure sheaf on an affine scheme
Spec A we have already seen that (!) x is the local ring of a prime ideal
XE Spec A. Owing to this the local properties considered in § 1 for affine

schemes, such as regularity of a point, tangent space, etc. automatically
carryover to arbitrary schemes.
Properties stated in § 1 in terms of a topological space such as irredu-
cibility, dimension, etc. are also applicable to an arbitrary scheme.
Finally, certain concepts introduced earlier for quasiprojective varieties
at once carryover to schemes. A rational morphism of a scheme X into
Y is a class of equivalent morphisms qJ : U -+ Y, where U is an open dense
set in X, and where morphisms qJ: U -+ Y and lP: V -+ Yare called
equivalent if they agree on Un V. Two schemes X and Yare called
birationally isomorphic if they have isomorphic open dense subsets
(see the proposition in Ch. I, §4.3).
2. Pasting of Schemes. By definition, every scheme is covered by

open sets isomorphic to affine schemes (or, as we say briefly, by affine
open sets). Can we perhaps recover the scheme X by knowing such a
U
covering X = U",? We consider this problem in a somewhat greater
generality, without assuming that the open sets U", are necessarily
affine.
First of all, we may remark that every open subset U of a scheme X is
a scheme; this follows from the fact that every point has an affine neigh-
bourhood V and that the D(f) C V from a basis of open sets.
U
If X = U", is an open covering, then the schemes U'" are not in-
dependent: U", and Up have the isomorphic open subset U",n Up.
Therefore we start out from the following data; a system of schemes
U "" oc E I, in each of them a system of open subsets U"',p C U"" OC, f3 E I,
U"','" = U "" and a system of isomorphisms of schemes qJ""p: U"',p -+ Up,,,,.
Let us find out how we can construct a scheme X, an open covering
X = U~, and a system of isomorphisms lP",: U '" -+~, such that lP",
restricted to U""p determines an isomorphism of the schemes U"',p and
~n Vp, and lPpqJ""plP; 1 is the identity mapping of ~n J.P. If such a
scheme X exists, then we say that it is obtained by a pasting of the
schemes U",.
For a pasting to be possible it is necessary, as is easy to verify, that
the following conditions hold:
qJ""", = 1, oc E I, qJ""pqJp,,,, = 1, OC, f3 E I. ( 1)
The restriction qJ~,p of the morphism qJ""p to U""p n U""y is an isomorphism

of U"',p n U"',y and Up,,,, n U P,Y' and these isomorphisms are connected
by the relations
(2)
§ 3. Schemes 247
rp:../3-__- -__
Ucc
The morphisms and schemes occurring in the conditions (1) and (2)
are illustrated in Fig. 15.
Next we show that if the conditions (1) and (2) are satisfied, then a
pasting is possible. First of all, we define X as a set. For this purpose we
introduce in T, the disjoint union of all the Ua , a relation by setting
x", y if x E Ua,p, yE U p.a' y= q>a,P(x). The conditions (1) and (2) guarantee
this relation is an equivalence relation. We denote by X the factor set of
this equivalence relation, and let p: T ~X be the canonical projection.
We introduce a topology in X by taking as open those sets U C X
for which p-l(U) is open. (The topology in T is determined by the
open sets Uvv", where vv" is open in U a.) It is easy to see that p establishes
a homeomorphism lPa of the sets U a with open subsets Va C X and
that X = Uv,..
Finally, we define the sheaf (!)x on X as follows. For a W contained
in some v,. we set (!)x(W) = (!)uJlP; l(W)), choosing an arbitrary v,.) W
The choice of another Vp ) W replaces (!)x(W) by an isomorphic ring.
The homomorphisms QU;" where W' eWe v,., are defined in the obvious
manner. Therefore the presheaf (!)x is defined not on all the sets, but
those W on which it is defined form a basis of open sets. The situation is
the same as in the definition of the structure sheaf on Spec A . In this way
we can extend the definition of (!)x(U) to all open sets U C X as the
projective limit (!)x(W), where We U are those open sets on which it
was previously defined. There remains a standard verification of a large
number of properties (that (!)x is a sheaf, X a scheme, etc.), which we
omit.
As a first application of this construction we define a scheme IP"(A),
a so-called projective space over the ring A. For this purpose we consider
n + 1 independent variables To, ... , 7;., and in the ring of fractions
A[To, ... , 7;.J(To ... Tn ) the subrings Ai = A [To/ T;, ... , 7;./ 1]. We set
U i = Spec Ab U ij = D(TIT;) CUi' By definition, U ij = Spec A ij , where
Aij = (Ai)(Tj/T;l consists of the elements
F(To, .. ·, 7;.)/ T; P 1? E A[To, .. ·, 7;.JTo ... Tn )
for which F is a form of degree p + q. Hence it follows that Aij and Aj;
agree in A[To, ... , T"J(To ... Tn)' and so we have the natural isomorphism
CPij: U;r~ Uji' It is easy to check that the conditions (1) and (2) both hold.
As a result of pasting we obtain the scheme IPn(A).
3. Closed Subschemes. If the ring homomorphism A: A ~ B is an

epimorphism, then the mapping a A: Spec B ~ Spec A determines a
homeomorphism of Spec Bonto the closed subset V(a) C Spec A, where a
is the kernel of A. In that case Spec B is called a closed subscheme of
Spec A, and the morphism a A a closed embedding. We generalize these
concepts at once to arbitrary schemes.
A morphisms of schemes cP: Y ~ X is called a closed embedding
(or closed immersion) if every point x E X has an affine neighbourhood U
such that the scheme cP -leU) is affine and the homomorphism lpu:{!}x(U)-+
(!}y(cP -leU)) is an epimorphism.
In that case Y is called a closed subscheme of X.
Since closedness is a local property, cp(Y) then is a closed subset of the
topological space X.
To make this definition consistent with the example from which we
have started out we prove the following proposition.
Proposition. If X is an affine scheme, X = Spec A, and cp: Y ~ X
a closed embedding, then Y is also affine: Y = Spec B, and cp = a A, where
A: A ~ B is an epimorphism of rings.
We can find a covering X = UU i, Ui=D(f;),fiEA, such that cp-l(U;)
= Spec Ai and lpi : A Ii ~ Ai is an epimorphism. We set ker lpi = ai C A Ii'
n
Q~i = a Ai, Ai- 1 ai = a. The morphism cp makes Y into a scheme over A.
But since a C Ai- 1 a;, under the action of A on {!} y( cp - 1 (U;)) the ideal a acts
trivially. In other words, Y is a scheme over A/a. This shows that the
diagram
y~ )X
Spec A/a
is commutative, where v is a closed embedding.

The proposition will be proved if we can verify that u is an isomor-
phism. Locally u is given by homomorphisms [in the sets cp-l(U i) and
v-leU;)]
where]; is the image of fi in A/a. It is sufficient to show that all the Ui

are isomorphisms.
§3. Schemes 249
That Ui is an epimorphism follows at once from the fact that a C A;- l ai.
To prove that it is a monomorphism we use the following remarks. The
ring (1)y(q>-I(U i n Uj)) can be described in two ways:
(1)y(q>-I(U; n U)) = (Ai)",;(l;fJ) = (Aj)",(ljfil " (1)
We consider the localization homomorphisms
A~:AA-~(Af;h;(fj)=A(f;fj) "
From (1) it follows at once that
A~ai = Aj aj , (2)
where, for example, A~ai is the ideal generated by the elements
Suppose that a E A determines an element of the kernel of the homo-

morphism Ui' Then Ai(a) E ai' By (2) it then follows that
A~Ai(a) E Aj aj .
The left-hand side is the image of a under the localization A --+ A(f,fj)
and is therefore equal to AjAia), and the elements on the right-hand side
are of the form Aj(a)jA/.fil Thus,
A!(AifJ' Aj(a) - a) = 0 .
Therefore
Aj(.fi)l+ m Aia) = A/.firaj E a j
for some m. So we see that
(3)
where I and m can be chosen to be the same for all j. The relation (3),
which has been proved for all j, shows thatfil+ma E a, that is (i;)l+ma = 0,
where a is the image of A/a. This means that a determines the zero
element in (A/a)!,. The proposition is now proved.
Closed subschemes provide us with new examples of schemes. Thus,
the closed subschemes of the scheme IP"(A) give us a new extensive type
of schemes over the ring A, generalizing projective varieties. Even our
customary quasiprojective varieties contain vastly more closed sub-
schemes than closed subvarieties.
For example, on the affine line X = Spec k[T] a closed subscheme
other than X is of the form Speck[T]/(F), where F(T) is an arbitrary
polynomial, whereas closed subvarieties correspond only to the collec-
tion of roots of these polynomials without reflecting the multiplicities
of the roots.
If cp : X -+ Y is a morphism of schemes and Y' a closed subscheme of

Y, then we can determine its inverse image cp - 1 (Y'), which is a closed
sub scheme of X. We only treat the case when X and Yare affine schemes,
X = Spec A, Y = Spec B, and cp = a A, A: B -+ A. Then a closed embedding
of Y' in Y is determined by the natural homomorphism B -+ Bjb. If
A(b)A = A, then the set cp - 1 (Y') is empty. If this is not so, then clearly
X' = Spec Aj A(b)A is a closed subscheme of X, which is called the
inverse image of Y'. As a topological space it is actually the inverse
image of the subspace Y' C Y.
For example, if X and Yare isomorphic to the affine line /AI over a
field k, if cp is given by the mapping cp(x) = x 2, and if the characteristic of
k is not 2, then cp -ley) for y;6 0 consists of two connected components
isomorphic to Speck (that is, two "ordinary" points), and cp -ley)
= Speck[TJj(T2) for y = O. This example shows that schemes with
nilpotent elements in rings (9(U) can arise in the most classical situations.
We have already had occasion to comment that it is natural to define the
inverse images of subvarieties of codimension 1 as divisors, that is, as
subvarieties with multiplicities. In the simplest cases these multiplicities
prove sufficient to specify a sheaf on these subvarieties. In the general
case this is a palliative: it is clear that the inverse image under a morphism
of two objects must be an object of the same kind, in our case a scheme,
but here we very often arrive at schemes with nilpotent elements.
Even more extreme is the situation in the example when X = Y = IA 1
is an affine line over a field k of characteristic p and cp(x) = x p • This
mapping is one-to-one, but not an isomorphism. Applying our concept
of inverse image we see that cp - 1 (x) = Spec k[TJj(TP), that is, the inverse
image of every point contains nilpotent elements in its sheaf. It is interest-
ing that in this case X and Yare algebraic groups under addition, and
cp is a homomorphism. Therefore it is natural to expect that cp - 1 (0) is
also a "group" of some new type. In the next section we shall see that this
IS so.
4. Reducibility and Nilpotents. If the rings (9x(U) have no nilpotent

elements, then the scheme X is called reduced. With every scheme X
there is associated a reduced closed subscheme X' whose topological
space coincides with X. For an open set U C X the ring (9x'(U) is defined
as the quotient ring of (9x(U) by its nil-radical (the ideal generated by all
nilpotent elements). This scheme is denoted by X red •
Let us consider some examples, with the object of illustrating the
role of non-reduced schemes in the study of "classical" quasiprojective
varieties. Let X be a quasi projective variety defined over a field k, and
Io=Speck. What are the morphisms cp:Io-+X (as a scheme over k)?
Since lois a quasiprojective variety consisting of a single point Yo and
§3. Schemes 251
since morphisms of quasiprojective varieties are the same thing as regular

mappings, ({J is completely specified by the image x = ((J(yo) of Yo' Of
these morphisms there are just as many as there are closed points
XEX.
Now we complicate this example somewhat by introducing nilpotent
elements.
We set II = Spec T 1 , where Tl =k[T]/(T 2 ), and we investigate the
morphisms ({J:ll-+X, Since 10 is a closed subscheme in II (IO=(Il)red)'
we see that ({J determines a morphism q/ : 10 -+ X specified by the point
x EX. It is sufficient for us to give the homomorphism lpu: lPx(U)-+ Tl
for affine open sets U C X. If x¢ U, then lpu =0. Let x E U. Since lPx(U)
= k + mx, where mx is the maximal ideal of the point x, we see that
lpu is uniquely determined by its action on mx and by lpu(mJ C ke,
where e is the image of Tin T 1 • As e2 =0, we have lpu(m;) =0, hence lpu
determines a linear mapping of mx/m; into kIl, that is, into k.
The space of linear functions on mx/m; is the tangent space ex to
X at x. Therefore, every morphism ({J: II -+ X for which ({J(I1 ) = x determines
a tangent vector at x. It is easy to verify that this establishes a one-to-one
correspondence between these morphisms and vectors in ex'
This result has the following geometrical interpretation. We consider
an affine neighbourhood U of x and in it a closed sub scheme
Tx = Spec k[U]/m;, where mx is the maximal ideal of x in k[U]. The
homomorphism k[U] -+k[U]/m; determines a closed embedding
Tx-+ U. It is easy to see that Yx is also a closed subscheme in X and does
not depend on the choice of the neighbourhood U. The arguments
above show that every morphism ({J:I 1 -+ X is of the form ({J :jlp,
where lp is a morphism II -+ Tx , and j a closed embedding of Yx in X.
Thus, the morphisms II -+ X carrying 10 into x, are in one-to-one cor-
respondence with the morphisms II -+ Yx. The scheme Yx is fairly large:
the tangent space to it at x is the same as that to X. But it is also small
enough for the morphism ({J: II -+ Yx to be uniquely determined by its
differential dxo({J, Xo = 10 , This is the geometrical interpretation of our
computations. It justifies the name "infinitely small neighbourhood oj x
of the first order" for the subscheme Yx. Similarly we can define an
"infinitely small neighbourhood of x of order n". These definitions lead to
a "theory of jets" on algebraic varieties.
As a second example we consider the concept of a family of sub-
varieties of a variety X, which was defined in Ch. I, §6.5, as a subvariety
r in the product X x S, where S is an algebraic variety. For a point
s E S the subvariety Y of X determined by Yx s = (X x s)nr is determined
the variety of this family corresponding to s.
Assuming now that S is an arbitrary scheme, the same definition can
be preserved verbatim. We only have to make use of the concept of
products of schemes, which will be defined in the next section. For ex-
ample, if S = 1 1 , then a family with base S is a subscheme r of X x 1 1 ,
It is easy to verify that if So is the closed point of 11 , then the embedding
So -+ I 1 determines the closed embedding X = X x So -+ X x I l' The
inverse image of r determines a subscheme rso in X -the unit sub-
scheme of the family. The concept of a family of divisors as defined in
Ch. III, §3.5, can also be generalized. Although the family at which we
arrive is purely "infinitesimal", its presence can be very important. The
subscheme (or divisor) rso can turn out to be reduced, that is, a quasi-
projective subvariety in the variety X, and the possibility of including it
in this family points to the existence of a subvariety of infinitely small
deformations. Every family in the previous sense containing the sub-
variety Y, of course, also determines an infinitely small deformation of
this subvariety. However, there exist, for example, a surface X and on it
a curve Y for which the set of infinitely small deformations is isomorphic
to k and which cannot be included in any family of curves other than
Y (that is, it has no finite deformations). An example is contained in the
book [24], Lecture 22. This example makes it particularly clear how
schemes with nilpotent elements naturally arise in connection even with
"classical" problems of the geometry of algebraic varieties.
5. Finiteness Conditions. Two properties of schemes, which we shall

analyse presently, have the character of "finite-dimensionality".
A scheme X is called Noetherian if it has a finite covering of affine
open sets
X=UU;, Ui=SpecA i , (1)
in which the rings Ai are Noetherian.

A scheme X over a ring B is called a scheme offinite type over B if it
has a finite covering (1) in which the Ai are algebras of finite type over B.
Obviously, a scheme of finite type over a Noetherian ring is Noe-
therian.
We shall prove propositions of one and the same kind referring to
both these concepts.
Proposition 1. If an affine scheme Spec A is Noetherian, then the ring
A is Noetherian.
By hypothesis, there exists a finite covering (1) such that the rings Ai
are Noetherian. Let a 1 C \12 C ... be a chain of ideals in A. As was shown
in §2.2, A = (9(X), where (9 is the structure sheaf on X. We consider the
ideals a~) = l!~ianAi C Ai' Since the rings Ai are Noetherian and there are
finitely many of them, there exists an N such that
(2)
§ 3. Schemes 253
for all i and all n;;;, N. Let us show that then an+1 = an for n;;;, N. Indeed,
since the Vi form a covering of X, it follows from (2) that
(e~an+ 1)(9 x = (e~an)(9 x
for all points x E X and all n;;;, N. It now remains to repeat the arguments
in §2.2. If U E an + l' then
U = ax/b x, ax E an' bx E A, bAx) i= 0 .
There exist points x 1, ... , Xr and elements c l' ... , C r E A such that
c 1 bx1 + ... + crb xr = 1. Then
that is, an = an+l'

Proposition 2. If an affine scheme Spec A has finite type over a ring B,
then A is an algebra of finite type over B.
By hypothesis, there exists a covering (1) such that the algebras Ai
have finite type over B. Since the space Spec Ai is compact, it has a finite
covering by principal open sets D(f), f EA. The corresponding algebras
(A;)I = A I have finite type over B. Therefore, we may assume at once
that in (1) Vi = D(f) Suppose that the generators of the algebra Ai over
U
B are of the form Xu/.ffii. On the other hand, as D(f;) = Spec A, there
exist elements gi E A such that
(3)
We denote by A' C A the subalgebra generated over B by the elements
Xij'/;' and gi' and we show that A' = A. Let x E A. By hypothesis, x E A Ii
for all/;. This means that there exists an n (by taking it sufficiently large
we may assume it to be independent of i) such that.fl'x belongs to the
subalgebra generated over B by the elements Xij and /;. In particular,
.fl'xEA' (4)
for all /;. By raising (3) to a sufficiently high power we obtain a relation
L.fl'g~n) = 1, where the g~n) belong to the subalgebra generated over B
by the elements/; and gj' In particular, g~n) E A'. Multiplying the relations
(4) by g~n) and adding up we see that x E A'.
Exercises
1. Let X be a ringed space and G a group consisting of automorphisms of X. Define
a set Y as the factor set of the points of X with respect to G, and let p : X --> Y be the natural
projection. Introduce in ¥the topology in which a set U C Yis open ifand only ifp-l (U) eX
is open. Finally, define a presheaf (!Jy by the condition: (!Jy(U) = (i)x(p-'(U))G Here A G
denotes the set of G-invariant elements of A [it must be checked that G is in a natural sense
a group of automorphisms of the ring (i)X(p-l(U))].
Show that (Y, lPy) is a ringed space. It is called the factor space of X with respect to G
and is denoted by X/G.
2. Let k be an infinite field, /A2 an affine plane over k, X = fA.2 - (0, 0), and let G
consist of the automorphisms (x, y)->(ocx, ocy), oc E k, oc =I o. Show that in the notation of
Exercise 1 the ringed space Y coincides with the projective line JPl over k.
3. Let X be the same as in Exercise 2, but let G consist of the automorphisms
(x,y)->(ocx,oc- 1 y),OCEk,oc=lO. Show that Y is a scheme. Show also that if X=fA.2, and
G the same as above, then Y is not a scheme.
4. Investigate the inverse images of the points x E Spec 7l in the morphism .cp in
Example 3 of § 1.1.
5. Investigate the inverse images of points under the morphism X -> Y projecting
the circle x 2 + y2 = 1 onto the x-axis :fix, y) = x, where all the varieties are defined over the
field IR of real numbers. In other words,
X = SpeclR[Tl' T2 ]/(Tl + Tl- I), Y = SpeclR[Tl].
6. Show that in Example 5 of§3.1 the points of the varieties coincide with the closed
points of the scheme X.
7. Let T be a homogeneous ring, T = EB Tn' Tn· Tm C Tn+m. An ideal aCT is called
n>O
homogeneous if a = EB (a n Tn). Denote by Proj T the collection of homogeneous prime
n>O
ideals peT that do not contain the ideal EB Tn' and introduce in this set the topology
n>O
induced by the embedding ProjTcSpecT. For a homogeneous elementfETm, m>O,
let I(J) denote the subring of Tr consisting of the quotients g/ f\ g E Tmk, k ~ O. Set
G+(f)= D(f)nProjT.
Let '!pr be the composition of mappings G +(f)->D(f)->SpecTr->SpecT(f). Show that '!pr
is a homeomorphism between G+ (f) and SpecT(f). Show that the structure sheaves over
Specl(J) (for all homogeneous f), carried over by means of '!pr to ProjT, determine a single
sheaf lp and that (ProjT, lP) is a scheme. This scheme is also denoted by ProjT.
8. Show that if in the notation of Exercise 7, T is a graded algebra over a ring A,
that is, Tn . A C Tn' then in this way a natural structure of a scheme over A is defined in the
scheme Proj T.
9. In the notation of Exercise 7, let T=A[To, ... , T,,] with the usual grading by
degrees. Show that the scheme Proj T is isomorphic to JPn(A).
10. Let Y be an affine n-dimensional variety over a field k, let y be a simple point of
it and my C keY] the corresponding maximal ideal. In the notation of Exercise 7, set
T= EB m;, m~ = k[lJ. Show that Proj T= X, where X is the variety obtained from Y by
n>O
the a=-process centred at y, and that the morphism (j: Proj T -> Spec keY] corresponding
to the a-process is determined by the natural algebra structure over keY] which exists
in T (see Exercise 8).
§ 4. Products of Schemes
1. Defmition of a Product. It would hopeless to define the product of
two schemes X and Y in terms of the set of pairs (x, y), x E X, Y E Y. For
when X = Y =JA 1 , we have X x Y =JA2 , and to the points X x Y there
correspond irreducible subvarieties of the plane JA 2 • Consequently,
among them there are all irreducible curves which, of course, cannot be
represented in the form of pairs (x, y). Therefore, first of all we must try to
clarify what properties we expect from a product of schemes, and then

we must tackle the problem whether a scheme with these properties
exists. This was the way in which we arrived in Ch. I at the definition of
the product of quasiprojective varieties.
We consider schemes over an arbitrary ring A. By definition, this
means a scheme X and a morphism X ~ Spec A. We even consider a
more general situation: a morphism of two arbitrary schemes X ~s.
Such an object is called a scheme over S. It is clear how to define a
morphism of schemes cp : X ~ Sand tp: Y ~ S over S; this is a morphism
f: X ~ Y for which cp = tp -J
If cp : X ~ Sand tp: Y ~ S are two schemes over S, then evidently
their product over S (which we denote by X Xs Y) must have projections
onto the factors, that is, two morphisms of schemes over S, namely
Px: X x s Y ~ X and py : X x s Y ~ Y in a commutative diagram.
Furthermore, it is natural to require universality of the product. This

means that for any scheme Z and morphisms u: Z ~ X and v: Z ~ Y for
which the diagram
commutes there exists a morphism h: Z ~ X Xs Y such that Pxh = u,

pyh = v, and the morphism h with these properties must be unique. The
morphism h is denoted by (u, v).
If a scheme X Xs Y satisfying these conditions exists, then evidently
it is unique to within an isomorphism. It is called the product of the
schemes X and Y over S. Occasionally, instead of using the term "scheme
X over S" one simply talks of the morphism cp : X ~ S, and then X x s Y is
called the fibred product of the morphisms cp and tp.
The definition is compatible with the definition of the product of

two objects of a category. In our case we consider the category of schemes
over S.
In the category of sets the fibred product of two mappings cp : X --+ S
and 1p : Y --+ S exists and coincides with the subset Z C X x Y consisting
of those pairs (x, y), x E X, Y E Y for which cp(x) = 1p(y). The situation is
similar in the category of quasiprojective varieties over an algebraically
closed field k.
The product of two schemes over a scheme S exists. The proof of this
assertion is essentially elementary, but somewhat lengthy. It can be found
in the book [15J, Ch. I. We confine ourselves to some remarks, which may
help the reader to reconstruct a proof.
If X, Y, and S are affine schemes, X = Spec A, Y= Spec B, S = Spec C,
then the specification of X and Y as schemes over S determines in A and
B algebra structures over C. In this case the scheme Z = Spec (A@cB)
is the product of X and Y over S if we endow it with the projections
Px = '1: Z --+ X and py = ag : Z --+ Y, corresponding to the homomorphisms
f:A--+A@cB, f(a)=a®1, and g:B--+A®cB, g(b) = 1®b. This pro-
position is a simple consequence of the definition ofthe tensor product.
In the general case we have to consider a covering S = UJtVx, X = UU ap,
Y = Uv;.y by affine sets such that cp(Uap ) C JtVx, 1p(v;.y) C Wa. Then
cp : U ap --+ JtVx and 1p: v;.y --+ JtVx are affine schemes over JtVx, and by the
preceding the products U ap X w" v;.y exist. It is not hard to verify that
these schemes satisfy the conditions (1) and (2) in § 3.2 (for a suitable choice
of open subsets and isomorphisms, which are easy to indicate), so that
they can be pasted together into a single scheme. After this we have to
define the projections of this scheme onto X and Y and verify that the
condition of universality holds.
From the definition of the product it easily follows that it is associative:
(X Xs Y) xsZ=X xs(YxsZ).
If S is an affine scheme: S = Spec A, then X X SpecA Y is also denoted
by X X A Y.
An arbitrary scheme can be regarded as a scheme over 7L. Therefore
the product of any two schemes is defined over 7L : X x z Y. It is simply
called the product of the schemes and is denoted by X x Y.
As a first application of the concept of a product we define the inverse
image of a closed subscheme (in §3.3 this definition had only been given
for affine schemes). If Y is a closed sub scheme of a scheme X, if j: Y --+ X
is a closed embedding, and if cp : X' --+ X is any morphism, then for the
scheme Y' = Y x x X' by definition the morphism j': Y x x X' --+ X'
exists. It is not hard to verify that j' is a closed embedding, so that Y'
is a closed subscheme of X'. It is called the inverse image of Y under the
morphism cpo It is easy to check that for the case of affine schemes this
definition is the same as that given earlier.
The advantage of the new definition consists in that it may be applied

to certain other situations. For example, let x be a point of a scheme X,
not necessarily closed. We set T = Spec k(x} and define a morphism
T --+ X by the fact that <p(T} = x and lPu(l'9(U}) = 0 if the open affine set U
does not contain x. But if x E U, U = Spec A, then x is a prime ideal of A,
and we define lPu as the natural homomorphism A --+ k(x} into the field of
fractions of A/x. The homomorphisms lPu automatically extend to all
open sets U C X and define a morphism <p : T --+ X.
If <p : X' --+ X is another morphism, then the scheme X' x x T is called
the inverse image of x or the fibre of the morphism <p over x. It has a
morphism X' x x T --+ T, that is, it is a scheme over k(x}. If the point x is
not closed, then its inverse image is not closed, in general.
2. Group Schemes. The concept of products makes it possible

to carryover to schemes the definition of an algebraic group. For this
purpose we have to reformulate the definition of an algebraic group
given in Ch. III, §3.2, so that we do not talk of points but only of mor-
phisms.
Let <p:X --+S be a scheme over S. A group law is defmed by a mor-
phism
fl:XXsX--+X.
The role of the unit element is played by a morphism
e:S--+X
such that <p . e = 1 (we have already seen several times that, for example,
for schemes over a field k the morphism Spec k--+ X determines a point
in X).
The association of every element with its inverse is replaced by the
specification of a morphism
i:X--+X.
The property of the unit element is expressed by the fact that
W (e<p, 1) = W (1, e<p) = 1 (1)
(1 is the identity morphism).
The property of the inverse element is expressed by the condition
W(i,1}=W(1,i}=e<p. (2)
It remains to write down the conditions of associativity. For this purpose
we observe that by the associativity of the product of schemes we have
two morphisms X x sX x sX --+ X x sX: (fl, 1) and (1, fl). Our condition
has the form
W(fl,1}=W(1,fl}· (3)
If the conditions (1), (2), and (3) are satisfied, then a scheme X over S
with the morphisms Jl, e and i is called a group scheme over S.
We leave it to the reader to make the natural definition of homo-
morphism and isomorphism of group schemes.
Here is a typical example, which shows the usefulness of extending the
concept of an algebraic group to that of a group scheme. Let X = Y = Ga
be the scheme of JAl over an algebraically closed field k of charac-
teristic p in which the group law is defined as Jl(x, y) = x + y. These are
even algebraic groups with which we are already acquainted. We consider
the homomorphism of these groups: f(x) = x p • As a point mapping it is
a monomorphism and as a mapping of abstract groups it is an iso-
morphism, but as a regular mapping of varieties it is not an isomorphism.
This is a serious difference from the usual situation in the theory of
groups.
In the preceding subsection we have seen that if we regard f as a
morphism of schemes, then the inverse image of every point is a non-
trivial scheme (that is, not Spec k). It is natural to try and turn f- 1 (0)
into a group scheme. For this purpose we denote the scheme by Z and
its closed embedding in X by j. We consider the morphism
The reader is advised to show as an exercise that there is a morphism
Jl' : Z xk Z - Z
such that Jl. U,j) = j. Jl' and that Jl' turns Z into a group scheme.
It can be shown that Z is the kernel of the homomorphism f in the
sense as this is understood in the theory of categories. Generally speaking,
the category of commutative algebraic groups over a field k of finite
characteristic is not an Abelian category, however, by extending it to
the category of commutative group schemes over k we arrive at an
Abelian category.
3. Separation. Finally we explain what is perhaps the most important

application of the concept of a product, namely to the question of separa-
tion of schemes.
The image of the morphism J=(1,1); X-XxsX is called the
diagonal.
A scheme X over S is called sepflrated if its diagonal is closed. A
scheme X is called separated if it is separated as a scheme over 7L.
The analogous condition for topological spaces defines Hausdorff
spaces ([8], § 8.1). In the case of schemes the meaning of this condition
is slightly different: the topological space connected with the scheme
is in any case almost never Hausdorff. To get a feeling for the meaning
of the condition of separation we give an example of a non-separated
scheme.
Let Uland U 2 be two copies of an affine line over a field k and let
U 12 C U 1, U 21 C U 2 be open sets obtaining by deleting a point ° (for
some fixed choice of coordinates T1 on Uland 12 on U 2). The mapping
q> that associates with every point x E U 12 the point u' E U 21 with the
same coordinate is, of course, an isomorphism. Evidently the conditions
necessary for pasting together Uland U 2 with respect to U 12 and U 21
are satisfied. As a result we obtain a scheme X over k, which is called
an affine line with ° as a branch point. In fact, it has two points 01 and 02
°
that are obtained from in Uland U 2, respectively. Let us show that
this scheme is not separated over k.
The closed points of the scheme X x k X are of the form (Xl' X 2 ),
where Xl and X2 are closed points in X, and the mapping LI is given by
LI(x)=(x,x). Since the scheme X, by construction, is covered by two
affine sets V1 and V2 isomorphic to Uland U 2, we see that X x X is
covered by the four sets V1 x V1, V1 X V2, V2 X V1, and V2 x V2.
Consider, for example, the set V1 x V2 • It is isomorphic to /A.. 1 x /A.. 1 ,
and its intersection with LI (X) consists of the points (x, x), XE V1 n V2 = V 12.
From this it is already clear that LI (X) is not closed, in fact, not even its
intersection with V1 x Vi is closed. To complete the picture we can com-
pute the closure of LI(X). The closure of LI(X)n (V1 x V2) in V1 x V2 is
evidently obtained by adding the point (° 1 ,°2 ). Considering in the
same way all the four open sets Vi x Vj, i,j= 1,2, we find that the closure
of LI(X) is obtained by adding two points: (° 1,°2) and (02,0 1). Hence
it follows that the closure of LI (X) is isomorphic to the line /A.. 1 in which
the point ° is "split" into four points: (° 1,°1), (02, 02), (°1,°2), (°2, 01)'
of which the first two belong to LI (X), but the second two do not.
To give a clearer idea of the influence of non-separation on properties
of a scheme we analyse this example of a scheme X in somewhat more
detail. The fields k(V1) and k(V2) are isomorphic and determine a field,
which we naturally call a field of rational functions on X. The local rings
(!) x of points XE X are subrings of this field. Let us find out what the rings
(!) 0, and (!) 02 are.
Clearly (!) 0, is the same as the local ring of 01 on V1. Since the iso-
morphism between U 12 and U 21 extends to the identity isomorphism
between Uland U 2, here the functions in (!) 0, correspond to functions
in (!) 02' and this means that (!) 0, = (!)02. Thus, two distinct points have one
and the same local ring. Furthermore, any function in this ring takes at
01 and 02 the same value: these two points are not separable by means of
rational functions. It can be shown that in the general case non-separa-
tion is connected with a similar phenomenon.
Let us now proceed to the general analysis of the concept of separa-

tion.
Proposition 1. An affine scheme X over a ring B is separated and
,1 : X ~ X x B X is a closed embedding.
Let X = Spec A and let A be a B-algebra. Since X X B X = Spec(A ®B A),
the morphism ,1: X ~ X X B X is associated with a homomorphism
A: A ®B A ~ A. According to the definition A is specified by the fact that
AU= 1, AV= 1 (1)
for homomorphisms u, v : A ~ A ®B A
u(a) = a® 1, v(a) = 1 ®a.
Hence it is easy to see that A(a®b) = abo From this, as in fact already
from (1), it follows that A is an epimorphism, and this means that ,1 is a
closed embedding.
Since every scheme is covered by affine sets, which are separated,
non-separation must be connected with certain properties of the pasting
of affine schemes. This is corroborated by the following result in which
we consider only the case when X is a scheme over an affine scheme
S=SpecB.
Proposition 2. Let X = UUa be an affine covering for which the follow-
ing conditions hold:
1) all the sets U an Up are affine and
2) the ring <'Dx(Uan Up) is generated by the rings eg:"up<'Dx(Ua:) and
eg~"up<'Dx(Up). Then the scheme X is separated over B.
Let u, v: X x B X ~ X be the standard morphisms of a product. Then
,1- 1 (u - 1 (U a) n v - 1 ( Up)) = ,1- 1 (u - 1 (U a)) n ,1- 1 (v - 1( Up)) = Ua n Up. (2)
On the other hand, from the definition of a product it follows easily that
for any open sets U, VeX the open set u- 1 (U)nv- 1 (V)eXxX is
isomorphic to U x V. Together with (2) this shows that for separation
of the scheme X it is sufficient that the restrictions ,1a,p of the morphism
,1 to Uan Up
,1a,p: UanUp~Ua.xB Up
should have a closed image. But by 1) Uan Upisaffine: Uan U p= SpecC",p'
and by 2) the corresponding ring homomorphism A"C8>BAp~Ca.,p,
U a = Spec Aa. is an epimorphism. But this means that ,1a.,p is a closed
embedding.
It is not hard to show that the converse is also true. We verify only
one perfectly obvious, but useful, part of it: in a separated scheme the
intersection of two open affine sets is affine. In fact,
If U and V are affine, then so is U x V, and if X is separated, then L1 is a

closed embedding, therefore U fl V is a closed subscheme of an affme
scheme. According to the proposition in §3.3, it is itself affme.
Now we turn our attention to an interesting feature of the criterion
stated in Proposition 2: it does not depend on the morphism X --+S.
Thus, the property of separation of a scheme X over an affine scheme
S does not depend on the choice of S and the morphism X --+ S. It could
have been formulated by regarding X, for example, as a scheme over Z.
An important application of Proposition 2 is the verification that the
projective space ]pn(A) over an arbitrary ring A is separated. In this
case IPn(A)= U Ui> U j =SpecA[To/7;, ... ,T,,/1i.]. Since
i=Ot·.·~n
Ujfl Uj = SpecA[To/T;, ... , Tn/T;](Tj/T;) '
this set is obviously affine. Now llJAUj fl Uj ) consists of elements

F (To, ... , Tn}/Tf T" where F E A [To, ... , Tn] is a form of degree p + q. The
rings eg:"Uj llJx(Uj) and e~"UjllJx(Uj) consist of elements F/Tf and
G/T" where F and G are forms of degree p and q, respectively. Evidently
they generate llJX(Ujfl Uj)'
It is easy to verify that in a separated scheme a closed subscheme and
an open subset are separated. Hence it follows that quasiprojective
varieties are separated.
We turn our attention to those properties of quasiprojective varieties
that are connected with their separation. Particularly often we have
used the fact that a regular mapping is uniquely determined by its
restriction to any open dense subset. The corresponding property of
schemes is closely connected with separation. Namely, if a scheme X is
separated, then for any scheme Y and morphisms f: Y --+ X, g: Y --+ X
the set Z C Y consisting of the points for whichf(y) = g(y) is closed. For
we have the morphisms if, g): Y --+ X x X, and Z is the inverse image of
the diagonal under this morphism.
This shows that only for separated schemes are rational morphisms
a natural generalization of morphisms. If a scheme X is not separated,
then two distinct morphisms Y --+ X may define one and the same rational
morphism.
Another frequently occurring property is that a regular mapping
has a closed graph. Iff: Y --+ X is a morphism of schemes, its graph is
defined as the image of the morphism (1 ,j): Y --+ Y x X. It is the inverse
image of the diagonal in X x X relative to the morphism
fx 1:YxX--+XxX(fx 1)= (fopy,Px)'
where py: Y x X --+ Y and Px: Y x X --+ X are the natural projections.
Thus, the graph of a morphism is closed if X is a separated scheme.
Exercises
1. Let X and Y be schemes over an algebraically closed field k. Show that the cor-
respondence u .... (px(u), py(u)) establishes a one-to-one correspondence between the closed
points of the scheme XxtY and the pairs (x, y) where x and yare closed points in X and
Y, respectively.
2. Find all the points of the scheme SpecCCxJRSpeccc, where CC and IR are the fields
of real and complex numbers.
3. Let X be an affine group scheme over an affine scheme S = Spec B, X = Spec A,
A an algebra over B. Show that the group law determines a homomorphism p.: A .... A ® BA,
the unit morphism determines a homomorphism e: A .... B, and the inverse element an
automorphism i: A .... A. Formulate condition (1), (2), and (3) of §4.2 in terms of these
homomorphisms.
4. Show that the kernel Y of the homomorphism G..... G.: x .... x P, as constructed in
§4.2, is an atrme group scheme, Y = Spec A, A = k[T]j(TP). Compute in this case all the
homomorphisms introduced in Exercise 3.
5. Consider the analogous homomorphism of the multiplicative groups Gm .... Gm'
x .... x p, and compute its kernel Y'. Show that the group schemes Y (see Exercise 4) and
Y' are not isomorphic.
6. Let k be a field of characteristic 2. Show that to within an isomorphism there
exist only two group schemes X = Spec A, where A = k[T]jT2, namely the schemes
Y and Y' (Exercises 4 and 5).
7. Show that the non-separated scheme considered in §4.3 coincides with the scheme
of Exercise 3 in §3.
8. Show that the scheme Proj r is always separated (Exercise 8 to §3).
Chapter VI. Varieties
§ 1. DefInition and Examples

1. Defmitions. In this chapter we consider schemes that are more
closely connected with quasi projective varieties. These schemes are
called algebraic varieties. It is precisely this concept that we arrive at in
trying to give an invariant definition of an algebraic variety.
Defmition. A variety over an algebraically closed field is a reduced
separated scheme of finite type over k .
. A morphism of varieties is a morphism as schemes over k.
A variety X which is an affine scheme is called an affine variety.
As we have seen in Ch. V, § 3, every quasiprojective variety
determines a scheme. This scheme is a variety, which we also call
quasi projective.
By definition, every variety X has a finite covering X = U;, U
where the Ui are affine varieties. From this it follows that X is of finite
dimension. If X is irreducible, then all the Ui are dense in X and
dim X = dim Ui • Furthermore, they all are birationally isomorphic,
because Ui (\ Uj is open and dense both in Ui and Uj . Therefore the
fields of rational functions k(Ui ) are isomorphic to each other and can
be identified. The field so obtained is called the field of rational
functions on X and is denoted by k(X). The dimension of X is
equal to the transcendence degree of k(X).
A closed point of a variety X that belongs to an affine open subset U
is also closed in U and is a point of the corresponding affine variety
with coordinates in k. There are plenty of such points on X.
Proposition. The closed points are dense in any closed subset.
Observe, first of all, that in an affine variety (and even in an affine
scheme) every non-empty closed subset has a closed point. For a
non-empty closed subset Z of the space Spec A is of the form
Spec B, where B is a quotient ring of A. Since every ring has a
maximal ideal, Z has a closed point.
If X is an arbitrary variety, Z C X a closed subset, and Z E Z, then
it is sufficient to show that Z (\ U contains a closed point for every
264 Chapter VI. Varieties
neighbourhood U of z. We may restrict ourselves to affine U because

they form a basis of all open sets. For affine U, by the preceding,
Z n U has a closed point. But here is a hidden danger: the point may be
closed in U, but not in X. This actually occurs, for example, in the case
of the set U = Spec @ - {x}, where @ is the local ring of a closed point x
of a curve. Fortunately, in the case of varieties all is well: if a point
z E X is closed in some neighbourhood U of it, then it is also closed
in X. This follows from the fact that closed points x of varieties are
characterized by the property k(x) = k. For a point x is closed in X if
and only if it is closed in all affine open sets containing it, and for an
affine variety the condition k(x) = k evidently characterizes the closed
points. The field k(x) depends only on the local ring of x, hence does
not change when X is replaced by an open set U containing x. This
proves the proposition.
Since a variety is a reduced scheme, the elements fE@x(U) are
uniquely characterized by their valuesf(x) E k(x) for all x E U. According
to the proposition they are characterized by their values at closed
points. But then k(x) = k, so that the elements f E @x(U) can be
interpreted as functions on the set of closed points with values in k.
If <p: X -+ Yis a morphism of varieties, x E X and y = <p(x), then the
homomorphism oflocal rings <p*: @y-+@x determines a field embedding
k(y)-+k(x). If x is a closed point, then k(x)=k, hence k(y)=k, that is,
y is also closed. Therefore the image of a closed point is closed. Thus,
by interpreting the elements f E @y(U) as functions on closed points we
can define a homomorphism lpu: @y (U)-+@X(<p-l(U)} by the condition
lpuCf)(x) = f(<p(x)}. In other words, the specification of the mapping
<p : X -+ Y and even of its restriction to the set of closed points
determines a morphism.
Of course, a variety X has masses of non-reduced closed sub-
schemes. But every closed subset Z C X can be made into a reduced
scheme, or as we shall say henceforth, into a closed subvariety. If X
is an affine variety, X = Spec A, Z = V(a), then we set Z = Spec A/N',
where N' consists of all the elements a E A, a power of which is
contained in a (the inverse image of the nil-radical of the ring A/a).
The general case is obtained by pasting together.
All this shows how close varieties are to quasi projective varieties.
Indeed, all the local concepts and properties analysed in Ch. II, the
concept of a simple point, the theorem that the set of singular points
is closed, the properties of normal varieties, are preserved verbatim
for algebraic varieties. The same applies to properties of divisors and
differential forms.
The only properties whose transfer to varieties is not obvious are
those that are connected with projectiveness. Let us clarify at once
what conditions replace projectiveness in the case of arbitrary

varieties.
The property of being projective is, of course, far from "abstract".
But there is one proposition at our disposal, namely Theorem 3 in
Ch. I, § 5, that gives an "intrinsic" characterization of projective
varieties. We recall the definition:
A variety X is called complete if for every variety Y the projection
morphism p : X x Y-+ Y carries closed sets into closed sets.
The basic properties of projective varieties: closure of the image,
absence of non-constant everywhere regular functions (@x(X) = k),
were derived from Theorem 3 in Ch. I, § 5.2, and are therefore true for
complete varieties. Note that in proving closure of the image we made
use of the fact that the graph is closed. As we have seen in Ch. V, § 3,
this follows from the fact that varieties are separated.
Among all properties of projective varieties proved in Ch. I-IV
there is only one in which projectiveness is used directly and not via an
application of Theorem 3 in Ch. I, § 5: this is the very important
Theorem 3 of Ch. II, § 3. Next we show how it generalizes to arbitrary
complete varieties.
Theorem 1. If X is a smooth irreducible variety and cp: X -+ Y a
rational morphism of it into a complete variety, then the set of points at
which cp is not defined is of codimension not less than 2.
Let Ve X be the set of points at which cp is defined, rq> the graph
of the morphism cp: V-+ Yin Vx Y, and Z its closure in X x Y, which we
regard as a closed subset of X x Y. The image of Z under the projection
p : X x Y-+ X is closed because Yis complete. Since p(Z)) V, we see that
p(Z) = X. The restriction p: Z -+ p(Z) is a birational isomorphism:
it is an isomorphism between rq> and V The theorem is a consequence
of the following result:
Lemma. If p : Z -+ X is an epimorphism and a birational isomorphism,
and if X is a smooth variety, then the set of points at which the rational
morphism p - 1 is not defined is of codimension not less than two.
For cp = qp-l, where q is the restriction to Z of the projection
X x Y-+ Y. Therefore cp is defined at those points at which p - 1 is
defined.
Proof of the Lemma. We assume that there exists a subvariety Te X
of co dimension 1 such that p - 1 is not defined at any of its points. By
replacing Z, X, and Tby affine open subsets we may assume that they
are affine and that Tep(Z) e X. Let Z eN' and let u 1 , ..• , Urn be
coordinates in JAm as elements of @z(Z). We consider a point t E T
and represent the rational functions (p -1)* (uJ in the form
(p-1)* (Ui) = g;/h,
where gl, ... , gm, hE (!)t are coprime. Then

h· (P-l)* (Ui) = gi, p*(h)' Ui = P*(gi)'
Therefore gJr)=O for all points rE T at which h(r) =0, and this
contradicts the fact that the elements g 1, ... , gm, hE (!) t are coprime.
The varieties we have introduced turn out to be so close in their
properties to quasiprojective varieties that the question arises: are
perhaps the two concepts identical? A little later we shall show in
§ 2.3 that this is not so: there exist varieties that cannot be embedded in
any projective space. However, what is much more important, owing
to its invariant, intrinsic character, the concept of a variety turns out to
be a vastly more flexible instrument. Many constructions can be carried
out very simply and naturally within the framework of this concept.
Post factum it can be shown sometimes that we need not go beyond
the framework of quasiprojective or projective varieties; however,
often this is only of secondary interest. In the following three subsections
we give some important examples of such constructions.
For the simplest example we can point to the definition of the
product of varieties. Within the context of varieties the definition is
very simple: the arguments of eh. V, § 4.1, simplify considerably if we
make use of the fact that the set of closed points of the variety X x Y is
of the form (x, y), where x and yare closed points in X and Y,
respectively (see Exercises 1 and 2). In eh. I, § 5, we had to spend a
considerable effort on this definition, because we had to convince
ourselves that the product of quasiprojective varieties is again
quasiprojective.
Another example we can consider here is the concept of normalization
of a variety. Let X be an irreducible variety, and K a finite extension
of the field k(X). We show that there exists a normal irreducible variety
and a morphism V K of it, V K : X~~X, such that k(X~) = K and that
the embedding vi(: k(X)~k(X~) coincides with the given embedding
k(X)~K. This variety is unique: for two normalizations X~ and
X~ there exists an isomorphism f: X~ ~ X~ such that the diagram
Xv
K~/.
f
K
X-v
X
commutes. The variety X~ is called the normalization of X in K.
The uniqueness is proved word-for-word as in eh. II, § 5.2, where
we considered the case K = k(X). To prove the existence we consider
an affine covering X = UUi . The integral closure Ai of the ring k[U;]
in K is a finitely generated algebra, as we have seen in eh. II, § 5.2.
Therefore, in the field K the normalization VK,i: U~,i~ Ui of the affine

variety Ui exists and is affine. From the uniqueness of the normaliza-
tion it follows that viJ(Uin Uj) and VK~j(Uin Uj) are isomorphic.
This makes it possible to paste the varieties U~,i into a single
scheme Xi:, which obviously is a reduced irreducible scheme of finite
type over k. Let us show that the scheme Xi: is separated.
We have to show that in Xi: x Xi: the diagonal is closed, and for
this purpose it is sufficient to verify that the diagonal is closed in a
neighbourhood of every point ~ E Xi: x Xi:. Suppose that the morphism
v x v : Xi: x Xi: ~ X x X carries ~ into '1 E X X X and that U' is an
affine neighbourhood of '1 such that (v x V)-l(U' )= V'is affine. Its
existence follows from that of the normalization in the affine case.
Since X is separated, the scheme U = LI n U' is closed in U', hence affine.
From this it follows that the scheme (v x V)-l(U) is also affine and a
fortiori so is its irreducible component V containing ~. Let (jv: Xi:~ Xi:
x Xi: and (j: X ~ X x X be the diagonal morphisms. We set W = w) - l(V)
= V-I (j -1 (U). So we obtain the commutative diagram
in which the morphism (jv corresponds to a finite regular mapping of

affine varieties. All the more this is true for the morphism (jv: W~ V
(a module that is finite over a ring is a fortiori finite over a larger ring).
Applying Theorem 4 of Ch. I, § 5 we see that (jV(W) = V, and this shows
that the diagonal is closed in the neighbourhood V' of ~.
Thus, the scheme Xi: is an irreducible variety, and as a trivial
verification shows, it is the required normalization.
So we see that in the context of arbitrary varieties the construction
of the normalization is trivial. The question remains whether the
normalization of a quasiprojective variety is quasiprojective. This is
true, but we shall not give the proof here, which naturally is based
on purely projective arguments. It can be found, for example, in [20]
Ch. V,§4.
In the case of curves we can repeat the proofs of Theorem 6 and 7 in
Ch. II, § 5.3. They show that the normalization (in its function field)
of any irreducible curve is quasiprojective, and for a complete curve
projective. In particular, if the curve is smooth, then it is quasiprojective.
In fact, this is true for arbitrary curves, but the proof is more
complicated, and we omit it here.
2. Vector Bundles. One of the most important constructions of

algebraic varieties having a typical non-projective character is a vector
bundle. We recall that the general concept of a fibration does not
differ at all from a morphism of varieties p: X ~ S or from that of a
variety over S. We are interested in fibrations whose fibres are vector
spaces. In formulating this concept we have to bear in mind that an
n-dimensional vector space over a field k has the natural structure of an
algebraic variety isomorphic to N.
Definition. A family of vector spaces is a fibration p: E ~ X in
which every fibre p - 1(x), X E X, has the structure of a vector space
over k(x), and the corresponding structure of an algebraic variety is the
same as the structure of p -l(X) as inverse image of the point x under
the morphism p.
The fibre p-l(X) is denoted by Ex.
A morphism f of a family p: E ~ X into a family q: F ~ X is a
morphism f: E ~ F for which the diagram
is commutative (in particular, f maps Ex into Fx) and the mapping

fx: Ex ~ F x is linear over k(x).
It is obvious how to define an isomorphism of families.
The simplest example of a family is the direct product E = X x V,
where V is a vector space over k, and p is the projection of X x V
onto X. This family and any family isomorphic to it are called trivial.
Example 1. Let Vand Wbe two linear spaces of dimension m and n,
respectively. Let us find the common form of a morphism of the
trivial families f: X x V~X x W We choose bases Vl' •.• , Vrn in V and
Wl' •.. , Wn in Wand denote the corresponding coordinates by ~l' •.. , ~rn
and '11, ... , '1n· The projections p: X x V~ V and q: X x W~ W de-
termine elements Xi = p* ~i E (!)x x v(X x V) and Yj = q* '1j E (!)x x w(X x W).
It is obvious that closed points (J( E X X V and fJ E X X Ware uniquely
determined by the values Xi((J() E k and Yj(fJ) E k. Therefore the morphism
f is uniquely determined by the elements f* Yj E (!) X x v(X x V).
The composition of the isomorphism X ~ X X Vi and the embedding
X x Vi ~ X X V determines a morphism CPi: X ~ X x V. We set
aij = cpr f*Yj E (!)x(X). Then
(1)
For it is sufficient to verify this equality at all closed points IX E X X V,

and there it follows at once from the definition of a morphism of
families (linearity of the mappingfx).
Conversely, every matrix (aij), aijE (9x(X), leads by means of (i)
to a morphism f: X x V~X x W. Clearly we obtain an isomorphism
if and only if m=n and the determinant det(ai) is an invertible element
of the ring (9x(X).
If p: E ~ X is a family of vector spaces, then for every open set
U C X the fibration p: p - 1 (U) ~ U is a family of vector spaces. It is
called the restriction of the family E to U and is denoted by Elu.
Defmition. A family of vector spaces p: E ~ X is called a vector
bundle if every point x E X has a neighbourhood U such that the
restriction of the family E to U is trivial.
Obviously the dimension of the fibre Ex of a vector bundle is a
locally constant function on X, in particular, is constant when X is
connected. In that case the number dim Ex is called the rank of the
bundle and is denoted by rk E.
Example 2. Let V be an (n + i)-dimensional vector space, and
]P" the projective space consisting of lines I C V. The line corresponding
to a point x E lP n is denoted by Ix. In ]pn x V we consider the set E
of pairs (x, v) for which x E lP" and v E V, v being a closed point of Ix.
Obviously, this is the set of closed points of some quasiprojective
subvariety of lP n x V, which we also denote by E. The projection
]pn x V~]pn determines a morphism p : E ~ lPn. We show that p : E ~ ]pn
is a vector bundle. We introduce in Va coordinate system (xo, ... , x n ),
which we also take to be a system of homogeneous coordinates for ]pn.
The restriction of E to the open set Aj x V #0 consists of the points
~=(XO:Xl:'" :x n); (Yo, ... , Yn»), ti = Xi/Xj' Yi= tiYj,
and the mapping ~~(tl' ... , tn), Yo) determines an isomorphism of

this family with Uo x k.
The rank of the bundle we have constructed is 1. The projection
lP n x V~ V determines a morphism q: E ~ V. The reader can easily
verify that this morphism is the same as the a-process at the point
O=(O, ... ,O)EVand q-l(O)=lP n xO.
We consider the vector bundle p: E~X and a morphismf: X' ~X.

The product E'=ExxX' has a morphism p':E'~X'. This mor-
phism determines a vector bundle. For Elu ~ U x V, U C X, and
if U'=f- 1 (U), then E'lu,=ExuU'~U'X. This bundle is also denoted
by f*(E). Evidently, rkf*(E) = rk E.
Example 3. Let X be a projective variety, f: X ~lPn a closed

embedding of it in a projective space, and p: E -+ IP n the bundle of
Example 2. Then f*(E) is a bundle over X of rank 1. Generally
speaking, it depends on the embedding f and is its most important
invariant.
Since a vector bundle is locally trivial, it is pasted together from
several trivial bundles. This leads to an effective method of constructing
bundles.
U
Let X = Ua be a covering for which the bundle p: E -+ X is
trivial on each of the Ua' We fix an isomorphism
CfJa: p-l(U,,}:'" Ua xV.
For the intersection Uan Up we have two isomorphisms:
CfJap = CfJalp- l(U~nU p)' CfJpa = CfJplp- l(U~nU p): p-l(Uan Up) -+ (U an Up) X V.
Therefore CfJPaCfJ;/ is an automorphism of the bundle (UanUp)x V.
We now use the result of Example 1. Choosing a basis in the
vector space V we write the automorphism CfJPaCfJai/ in the form of a
matrix Ca,p with coefficients in (!)x(Uan Up). Obviously these matrices
satisfy the pasting conditions:
Ca,a = 1 , Ca,y = Ca,pCp,y on Uan Upn Uy . (2)
Conversely, the specification for arbitrary IX and P of a matrix Ca,p
with elements in (!)x(Uan Up) determines a vector bundle provided that
the matrices satisfy the conditions (2).
The matrices Ca,p are called the transition matrices of the bundle.
It is easy to clarify the dependence of the matrices Ca,p on the
choice of the isomorphisms CfJa' Other isomorphisms CfJ~ have the form
CfJ~ = fa CfJa , where !a is an automorphism of the trivial bundle Ua x V
The automorphism fa can be described by a matrix Ba with
coefficients in (!)x(Ua) having an inverse of the same form. So we
arrive at the new matrices
C~,p = BaCa,pBi 1 .
Conversely, every such change of the matrices Ca,p leads to an
isomorphic bundle.
3. Bundles and Sheaves. Vector bundles are generalizations of
vector spaces. Now we introduce an analogue of the points of a vector
space.
Defmition. A section of a vector bundle p: E ~ X is a morphism
s: X -+ E for which p . s = 1 on X.
In particular, s(x) = Ox (the null vector of EJ is a section, the
so-called null section.
§ 1. Defmition and Examples 271
The set of sections of a bundle E is denoted by 2 (E).
Example 1. A section f of a trivial bundle X x k- X of rank is simply

a morphism of X into IA\ that is, fE (!)x(X). Thus, 2(X x k)=(!)x(X).
In particular, 2(1Pn x k)';" k, and similarly 2(1Pn x V) = V.
Consider the bundle E of Example 2 above. Each of its sections
s:lPn_E determines, in particular, a section s:lPn_lPn x V, hence by
Cor. 2, Ch. I, § 5.2, has the form s(x) = (x, v) for some fIXed VE V. But since
S(X)E E, we see that VE (. for all XE lPn, consequently v=O. Thus, 2(E) =0.
This shows, in particular, that E is not isomorphic to the trivial bundle.
In terms of transition matrices a section s is given by associating
with every set U« a vector s« = (f«,1, .. .'/«,n),1:.,i E (!)x(UJ, with s« = C«,psp
in UlI.nUp.
Starting from the definition of a vector bundle it is easy to verify
that for two sections S1 and S2 there exists a section S1 + S2 such that
(S1 + S2) (x) = S1 (x) + S2(X)
for every point x E X. The sum on the right-hand side has a meaning
because S1 (x), S2(X) E Ex, and Ex is a vector space.
Similarly the equation
(f. s) (x) = f(x) s(x)
determines a product of a section with an element f E (!)x(X).
Thus, the set 2(E) is a module over (!)x(X). Let us associate
with every open set U C X the collection 2 (E, U) of sections of the
bundle E restricted to U. An obvious verification shows that we
obtain a sheaf, which is denoted by 2 E' This is a sheaf of Abelian
groups, but it also has a more delicate structure, which we shall now
determine in the general case.
DermitiOD. Let iF be a sheaf of Abelian groups and .t:§ a sheaf of
rings on a topological space X; suppose also that every open set
U C X determines a modu!e structure over t:§ (U) in iF (U). Under
these conditions fF is called a sheaf of modules over t:§ if the
multiplication iF(U)® t:§(U)-fF(U) commutes with the restriction
homomorphisms Q~, that is, if the diagram
fF (V) ® t:§ (V) ----+, fF (V)
/!~..F®/!~." 1 11!~·.F
fF(U) ® t:§(U) , iF(U)
commutes for U C V. In that case every fibre fF x of the sheaf iF is a

module over the fibre t:§x of t:§.
A homomorphism of two sheaves of modules :!F' and :!F" over one and
the same sheaf of rings C§ is a system of homomorphisms
IPu::!F'(U)-+:!F"(U) of modules over C§(U) for which the diagram
,:r _" 't:::

'l'v
:!F'(U) ------'----+) :!F"(U)

commutes for all U C V.
Clearly, the sheaf .::e E corresponding to the bundle is a sheaf of
modules over the structure sheaf (!)x.
Every operation that is invariantly defined over modules can be
extended to sheaves of modules. In particular, for arbitrary modules
over a ring A the following operations are defined:
MtBMl' M@AMl, M* = Hom (M, A), A~M.
Applying them to modules :!F(U) and :!Fl(U) over rings, and taking the
associated sheaves, we arrive at the sheaves :!FtB:!F1 , :!F@,&:!Fu :!F*,
A~:!F, which are called, respectively, the direct sum, the tensor product,
the dual sheaf, and the exterior power.
The sheaf of a trivial bundle of rank n is given by the fact that
.::e E(U) = (!)(ut, that is, .::e E is the direct sum of n copies of (!)x. Such a
sheaf is called free of rank n. Let :!F be a sheaf of modules over the
structure sheaf (!). If each point has a neighbourhood U such that the
sheaf :!Flu is free and of finite rank, then :!F is called a locally free
sheaf of finite rank. Evidently, if a sheaf :!F is locally free, then each
of its fibres :!F x is a free (!) x-module. The sheaf .::e E corresponding to
any vector bundle E is locally free of finite rank, because E is locally
isomorphic to a trivial bundle.
Theorem 2. The association E --+.::e E establishes a one-to-one cor-
respondence between vector bundles and locally free sheaves of finite
rank (both considered to within an isomorphism).
We now show how to associate a vector bundle to a locally free
sheaf :!F of finite rank. Clearly X can be assumed to be connected.
Suppose that X = U Ua is a covering such that :!Flu~ is a free sheaf and
IPa::!Flu~ ~(!)~"" the corresponding isomorphism. Then
(1)
is an isomorphism of sheaves of modules. Since X is connected, it
follows that all the numbers na are equal. We set na = n. Every
endomorphism of the sheaf of modules (!}'U is given by a matrix

C = (Cij), Cij E (!}u(U). Thus, the isomorphism (1) determines a matrix
C",p, and it is evident that these matrices satisfy the relations (2)
in § 1.2. Therefore they determine a vector bundle E. A trivial
verification, which we omit, shows that filE =:F. This proves the
theorem.
It is easy to check that the correspondence E -+ fil E between vector
bundles and locally free sheaves enables us to associate with every
homomorphism of bundles a homomorphism of sheaves of modules
over .(!}x. In other words, this is an equivalence of the two catagories.
We observe that the fibre of the bundle and the stalk of the
corresponding sheaf are totally different objects. For example, if
E = X x k, then filE = (!) x' Ex = k, and (fil E)x = (!) x' In the general case
the fibre Ex can be recovered from the stalk (fil E)x by means of the
relations
(2)
where mx is the maximal ideal of (!) x' It is sufficient to verify this

locally, assuming that E = U x k', filE = (!}'U, and then it is obvious.
Theorem 2 yields a convenient method of constructing bundles.
Example 2. Let E and F be vector bundles, fil E and fil F the
corresponding locally free sheaves. It is clear that the sheaves fil EEt>fil F,
fil E®fil F, fil~, A~fil E are locally free. Their corresponding bundles
are denoted by EEt>F, E®F, E*, APE. In the case p=rkE, APE is
denoted by det E.
If the bundles E and F in the covering X = UU" are defined by the
matrices C",p and D",p, then the bundles EEt>F, E®F, E* and APE
are given in the same covering by the matrices
(3)
For p = rk E the bundle APE is given by the one-dimensional matrices

detC",p.
From the relations (2) it follows that under these operations on the
bundles the corresponding operation on vector spaces is performed in
every fibre.
Example 3. Let X be a smooth variety. By assigning to an open set U
the group QP[U] of differentiable forms regular on U we evidently
determine a sheaf of modules over (!Jx. It is called the sheaf of p-dimen-
sional differential forms.
Theorem 2 of Ch. III, § 4.3, asserts that this sheaf is locally free.
Consequently, by Theorem 2 it determines a bundle, which is denoted
by QP. In particular, Ql is called the cotangent bundle.
The stalk of the sheaf IF of one-dimensional differential forms at a
point x E X is of the form ~ = (!}x dtl + ... + (!}x dt m where t 1 , ••• , tn are
local parameters at x, and the sum is direct. The homomorphism
IFx-~/mxIFx can be written in the form
from which it follows by (2) that
Q! ~ IFjmxIFx ~ mjm; . (4)
Obviously APQl = QP, detQl = an, n = dimX.

Example 4. The bundle dual to the cotangent bundle is called the
tangent bundle and is denoted by EJ. By virtue of (4), for every point
XEX
EJx=(mjm;)* ,
that is, it is the tangent space at x.
The last general question we wish to discuss in connection with
vector bundles are the concepts of subbundle and factor bundle.
Defmition.If a morphism of bundles <p : F _ E is a closed embedding
of varieties, then it is called an embedding of the bundles. In this case
<p(F) is called a subbundle of E.
Proposition. A sub bundle FeE of a vector bundle is locally a direct
summand.
The assertion is that every point x E X has a neighbourhood U and
a fibration Gover U such that
Elu ~FluEt> G. (5)
By assumption we can suppose that both F and G are trivial on U,
so E~UxW, F~UxV and f:U+V-UxW is a subbundle. Let
Xl' ... 'Xm be coordinates in V and Yl, ... ,Yn in W. Since f is a closed
embedding the associated f*: (!}u x w-(!}u x v is an epimorphism. Let
(a;), a;jE (!}(U), be the associated matrix. There is a matrix (bj/J, bjk E (!}(U),
such that
n
X k= L bjd*Yj, k= 1, ... , n ..
j= 1
The matrix bjk defines a map g: U x W - U x V. Since (aij)(b jk) = 1,

the composition g f: U x V - U x V is the identity. This shows that the
associated sheaves of modules satisfy the relation

2Elu~2Flu$ff ,
where ff is a free sheaf of (!)u modules.

Now we can define the factor bundle ElF with respect to a sub-
bundle FeE.
As a set, of course,
xeX
To introduce in it the structure of a variety we consider an open set U

on which (5) holds, and we identify U
ExlFx with an algebraic variety G.
xeU
It is easy to verify that these structures are compatible on distinct open
sets U and define ElF as a vector bundle.
The translation into the language of transition matrices is obvious.
If we take a covering X = UU" so that (5) is true for all U,,' then the
matrices C", p defining E can be written in the form
_ (V",p 0 )
C",p - * V"
",p
where V",p determines the bundle F, and V~,p the bundle ElF. Hence it
follows at once that
detE = detF detEIF. (6)
Example 5. Let X be a smooth variety, and Y e X a smooth closed
subvariety. We define the normal bundle Nx1y to Yin X. The definition
given in differential geometry is not applicable in the algebraic situation,
because it is connected with the notion of the orthogonal complement
W-L of a linear subspace We V. However, the space W-L is defined so
that it is isomorphic to V IW, and this we can utilize.
We denote by e~ the restriction of the bundle eX to the sub-
variety Y. It is defined as j*e x , where j: Y -X is a closed embedding.
The bundle e y is a subbundle of e~. For by definition e~ = j*e x
= j*((Q})*) = U*Q})*. The restrictions of differential forms from X to Y
determines a homomorphism cp :j*Q}-Q}, and
cp*: ey=(Q})*-U*Q})* = ex·
By definition,
Nx1y = e~/ey.
Let us compute the transition matrices of the normal bundle. The

homomorphism e~-NxlY determines a homomorphism
1p: Ntly-j*Q}
of the dual bundles. It is easy to see that 1p determines a closed

embedding, so that Ni,y can be regarded as a subbundle in j* Q}, and
that Q} is a factor with respect to this bundle. It is enough to verify
these assertions on open sets, on which our bundles are trivial, and then
they are obvious.
As we have seen, the forms du 1 , ..• , dUn are a basis of the
{9x(U)-module Q}[U], when the functions Ul' ... , Un determine local
parameters at an arbitrary point x E U. This basis defines a basis 1]1' •.• , I]n
of the {9y(U n Y)-module givell by the sheaf corresponding to the bundle
j*Qi. Here cp(I]J is the restriction of the form dUi to Y.
According to Theorem 5 of Ch. II, § 3, we can choose functions
U1 , ••• , Un such that U1 , •.• , Urn are local equations of Yin U. According
to the same theorem the restrictions of the forms du m+ 1 , ... , dUn
determine a basis in Q~[U n Y], hence '11' ... , '1m is a basis of the
(9y(U n Y)-module Ni,Y(U n Y).
Suppose that systems Ua,l, ... ,ua,n and Up,l' ... ,up,n are chosen in
the sets Ua and Up in the manner indicated. The transition matrix for
the bundle Qi is determined by the expansion
n
du a, i = L Ci,j dUp,j'
j=i
i = 1, ... , n, Ci,j E (9X(U) , (7)
and the transition matrix for j* Qi in the basis I] 1, ... , '1n is obtained by
restricting the elements of this matrix to Un Y.
Since Ua,i E (Up, l' ... , Up,m), i = 1, ... , m, on Uan Up, we have
m
Ua,i= L /;,jUp,j' i=1, ... ,m, /;'jE{9x(UanUp).
j= 1
Hence
m m
.dua, i = L fi,j dUp,j + L Up,j d/;,j. (8)
j=1 j=1
To make these formulae compatible with (7) we have to express d/;,j in

terms of du 1 , ... , dUn. But we are interested in the formulae for '1i that
are obtained by restricting all the functions occuring in it to Y. Since
Up,j = 0 on Y, j = 1, ... , m, the second group of terms in (8) disappears.
Thus,
m
'1a,i= L ];,j'1p,j' i= 1, ... ,m,
j= 1
where ];,j is the restriction of /;,j to Uan Upn Y. As we have seen, these
are the transition matrices of the bundle Ni,y. The matrices for Nx1y
are obtained by transposition and inversion. The transition to the
inverse matrix is equivalent to an interchange of the order of a and [3.
Finally, we obtain the simple formula

C",p=(h;)y) (9)
if
Up,j = L hi,ju", i in U,," Up .
Almost all constructions of this subsection are based fundamentally
on the possibility of specifying the bundle abstractly, without embedding
in a projective space. It can be shown, however, that a vector bundle
over a quasi projective variety is itself quasiprojective. We do not prove
this here.
4. Divisors and Line Bundles. To every divisor D on an irreducible

variety X there corresponds a linear space 2(D) (we do not assume X
to be smooth and we consider locally principal divisors). This association
can be turned into a sheaf on the variety X. For this purpose we
observe that a divisor D on X determines a divisor on any open subset
U eX: we have to restrict the local equations of the divisor D to U.
We denote the divisors so obtained by Du and we set
2 D(U) = 2(U, Du),
where 2(U, Du) is the space associated with the divisor Du on the variety
U. Obviously 2D(U) c k(X), and 2 D(V) C 2D(U) if U C V. We denote
the inclusion of 2 D(V) in 2D(U) by (I~.The system {2D(U), (I~}
determines a presheaf, in fact, a sheaf, as is easy to verify, which we
denote by 2 D •
Multiplication of elements fE2 D(U) by hE (9x(U) turns 2D into a
sheaf of modules over (9x. This sheaf is locally free. For if D is defined
on an open set U" by the local equation fa' then the elements g E 2 D (U,,)
are characterized by the condition gf" E (9x(Uo). This shows that the
mapping g ~ g f" determines an isomorphism
qJ,,:2DIUa--->(9xIUa' (1)
In § 1.3 we have seen that this sheaf determines a vector bundle ED' and
from (1) it follows that rk ED = 1. Bundles of rank 1 are called line bundles
(their fibres are straight lines). Let us write down the transition functions
of the bundle ED' Since the isomorphism (1) is given on U" by
multiplication by f", the automorphism qJpqJ; 1 is given on U,," Up by
multiplication by f,,-lfp' Note that f"-lfpE(9x(U",,Up) in view of the
consistency of the system f". Similarly (fa- 1 fp)-l = f p- 1fa E (9X(Ua" Up).
Thus, in this case the transition matrix qJ",p of order one can be written
in the form
(2)
If the divisor D is replaced by an equivalent divisor D' = D + (f),

f E k(X), then multiplication by f determines an isomorphism of the
modules fE(U, Du) and fE(U, D~). We have verified this in ChI III, § 1.5.
Evidently we obtain in this wayan isomorphism of the sheaves fED
and fED" The bundles ED and ED' even have identical transition
matrices. Thus, both the sheaf fED and the bundle ED correspond to an
integral class of divisors.
Theorem 3. The association D -+ fED -+ ED determines a one-to-one
correspondence between 1) divisor classes, 2) classes (to within iso-
morphism) of sheaves of (!)x-modules, -locally isomorphic to (!)x, and
3) classes of vector bundles of rank 1.
The correspondence between the sets 2) and 3) was established in
Theorem 2. Therefore it is enough for us to show that D-+ED
determines a one-to-one correspondence between the sets 1) and 3).
To show this we construct the inverse mapping.
Suppose that in the covering X = UU a the line bundle E is given
by transition matrices ({Ja,p of order one, where ({Ja,p E (!)x(Uan Up),
({J:'~ E (!)x(Uan Up). From the relations (2) in § 1.2 it follows that
({Jp,ff. = ({Ja,p
-1
an d
m -m- 1 m
Ya,p-yy,aYy,p on (3)
We fix an index y, which we denote by 0, and we set y =0 in (3). The

embedding (!)x(UanUp)-+k(X) enables us to regard the ({Ja,p as elements
of k(X), and (3) holds for them, as before. We set fa = ({JO,a' The system
of elements fa on the sets Ua is consistent, since
(4)
therefore determines a divisor D. Comparison of (2) and (4) shows that
E=E D·
We now prove that the divisor class D depends only on the bundle E
and not on the choice of the covering and the transition matrices ({Ja,p.
Two systems {({Ja,p, U,,;} and {({J~'Il' U~} can be compared on the covering
{UanU~} by setting (Pa,P,).,1l = ({Ja,p, (P~,f3').'Il=({J~'1l on UanUpnU~nU~.
Therefore we may assume from the very beginning that the covering is
U
common to the two cases: X = Ua . As was shown in § 1.2, we then have
(5)
By definition of the functions fa and f: ,

f: = !Po 1 ({JOa!Pa = !Po 1 fa!Pa ,
and by (5), D' = D - (!Po).
So we have actually constructed a mapping of the set 3) into 1). An

obvious substitution shows that it is inverse to the mapping D~ED'
This proves the theorem.
For any morphism f: X ~ Y the following relation holds:
(6)
whose simple verification is left to the reader.
The class of divisors corresponding by Theorem 3 to the line bundle
E is called its characteristic class and is denoted by c(E).
Example 1. If dim X = n, and Qn is the bundle introduced in § 1.2,
then c(Q") = K is the canonical class.
Example 2. Let X be a smooth variety and Y C X a smooth hyper-
surface. In this case the normal bundle Nx1y is linear. Let us compute
its characteristic class.
Suppose that Y is given in an affine covering X = UU" by local
equations f". Then f"-lfp=f,,,p, where f",p, f,,:)E{9(U rx nUp). Ac-
cording to the formulae (9) of § 1.3, the transition matrices of the bundle
N x /y have the formfrx,ply=(f,,-l fp)ly. But we have just seen thatf,,-l fp
are the transition matrices for the bundle E y . So we have proved the
formula
By (6) it then follows that

c(Nx1y) = Qy(Cy) ,
where Cy is the divisor class on X containing Y, and Qy: CI(X)~CI(Y)
is the homomorphism of restriction to Y. We recall the explicit way of
obtaining Qy: we have to replace Y by an equivalent divisor Y' not
containing Y as a component, and then restrict Y' to Y.
Since the divisor classes form a group, the correspondence
established in Theorem 3 determines a group operation also on the set
of line bundles or sheaves locally isomorphic to {9. From (2) it is clear
that addition of divisors corresponds to multiplication of one-
dimensional transition matrices. In a more invariant form this operation
is given as the tensor product of bundles or sheaves (see Theorem 2).
Here multiplication of sheaves by {9 plays the role of the unit element,
and the inverse to the sheaf .PD is .P _D' Therefore locally free sheaves of
{9-modules of rank 1 are also called invertible sheaves.
Although invertible sheaves and divisor classes correspond to each
other in a one-to-one manner, the former are technically more con-
venient to use. For example, the inverse image f* ff is defined for every
morphism f and every sheaf ff. It is easy to verify that if a sheaf ff is
invertible, then so is f* ff. The corresponding operation on divisor classes
requires for its definition arguments connected with a shift of the

support of a divisor.
The technical advantages of invertible sheaves are connected with a
fundamental phenomenon: in a closely related situation in the theory of
complex analytic manifolds the concepts of an invertible sheaf and a
divisor class are already inequivalent, and invertible sheaves give more
information and lead to more natural problems. On this point see
Exercises 6, 7, and 8 to Ch. VIII, § 2.
As an application of the preceding concepts we derive a relation
which we have stated and used in Ch. IV, § 2.3.
Theorem 4. The genus gy of a smooth curve Y on a smooth complete
swface X can be expressed by the formula
gy=!(Y+K, Y)+1, (7)
where K is the canonical class of X.

For a smooth subvariety Y C X in a smooth variety X formula (6)
of § 1.3 gives
From formula (3) of § 1.3 it follows that det(E*) = (detE)-l for every
bundle. Since
we obtain
Qy(c(Q~)) = c(Q'Y) - c(detNxlY) '
where dimX = 11, dim Y = m. Now let m = 11 - 1. We make use of the

results obtained in the discussion of Examples 1 and 2 and arrive at the
relation
(8)
Finally, if n = 2, m = 1, then the equality of the degrees of the divisors

on the two sides of the equation follows from (8).
Recalling that degQy(D) = (Y, D) and that by the Riemann-Roch
theorem degKy = 2gy - 2, we obtain
and the theorem follows.

Exercises
1. Let k be an algebraically closed field. We define a pseudovariety over k as a
ringed space in which every point has a neighbourhood isomorphic to Specm A,
where A is an algebra over k of finite type and without nilpotent elements', and where
the topology and the sheaf on Specm A are defined word for word as in Ch. V. Show that
by associating with every variety the set of its closed points we determine an isomorphism
of the categories of varieties and pseudovarieties.
2. Define the product of two pseudovarieties X and :y starting out from the fact
that X x Y consists of pairs (x, y), x E X, Y E Y, constructing an affine covering of this set
from affine coverings of X and Y, and using the definition of the product of affine
varieties in Ch. I.
3. Prove that a variety is complete if and only if its irreducible components are
complete.
4. A fibering X .... S (not necessarily a vector bundle!) is called locally trivial if
every point S E S has a neighbourhood U such that the restriction of X to U is
isomorphic to F x U (as a scheme over U). Show that if the basis of S and the fibre of a
locally trivial fibering X are complete, then so is X.
5. Determine the transition matrices of the bundle in Example 2 of § 1.2
corresponding to a covering of lP n by the sets !Ai. Find the characteristic class of this
bundle.
6. Let D be a divisor on a variety X for which the space !l' (D) is finite dimensional,
ff = ffD its corresponding invertible sheaf, and f a rational mapping in lPn, n = /(D) - 1,
that is associated with !l' (D) in accordance with Ch. III, § 1.5. Show that f is regular at
those and only those points x E X for which the stalk ff x is generated over (!) x
by the space I!x!l' (D).
7. Let X be a smooth affine variety: X = SpecA. Show that the module E>AX)
over A is isomorphic to the module of derivations of A, that is, the k-linear mappings
d: A .... A for which d(xy) = d(x) . y+x' d(y), x, yE A.
8. Show that the normal bundle to a line C in lP n is a sum of n - 1 isomorphic
I-dimensional bundles E. Find c(E).
9. Suppose that n-l hypersurfaces C I , ••• , Cn - l of degree ml"'" m n - l in lP n
intersect transversally with respect to a curve X. Find its genus.
10. Let f: E .... X be a vector bundle and X = UU. a covering over the elements of
which E is trivial: Elu."",U.xk". We embed k" in lP n as points with xo#O and paste
together the varieties U. x lP n by means of the transition matrices C•. P of E, which are
now regarded as matrices of projective transformations in lPn. Show that we can obtain
in a suitable manner a variety Ein which E is an open subset such that Eis smooth,!: E..... X
is regular, and its fibre is isomorphic to lPn.
11. In the notation of Exercise 10, let X = lPI, En the bundle ofrank 1 corresponding
to the divisor nxoo on lPI, n > O. Show that En - En is a curve which f maps
isomorphically onto lPI. Let Co be the null section of En' which evidentally is
contained in En' and F the fibre of En. Show that on the surface En: Co - Coo - nF.
Find (C~) and (C~).
12. Show that in the notation of Exercise II the restriction of divisors DE Div En to
the common fibre determines a homomorphism Cl E..... Z whose kernel is Z· F. Show
that Cl En is a free group with the two generators Co and F.
13. In the notation of Exercises 10-12 find the canonical class of the surface En'
14. Show that the surfaces En corresponding to distinct n;;. 0 are non-isomorphic.
Hint: Show that on En there is a unique irreducible curve with a negative square
and that this square is - n.
§ 2. Abstract and Quasiprojective Varieties

1. Chow's Lemma. We prove a result which throws some light on
the connections between complete and projective varieties. Of course,
every irreducible variety is birationally isomorphic to a projective
variety, for example, the projective closure of any of its open affine
subsets. However, in this direction one can prove significantly more:
Chow's Lemma. For every complete irreducible variety X there
exists a projective variety X and an epimorphism f: X - X that is a
birational isomorphism.
The idea of the proof is the same as that which we used to construct
a projective embedding of a normalization of a curve.
U
Let X = Ui be a finite affine covering. For every affine variety
Ui C JR' we denote by Yi its closure in the projective space lP n, ) JR'.
Obviously, the variety Y = II Yi is projective. .
We set U=nUi. The embeddings 1p: U-X and 1pi: U-UiC Yi
determine a morphism
cp : U - X x y, cp = 1p X II1pi .
We denote by X the closure of the set cp(U) in X x Y.

The projection Px: X x Y - X determines a morphism f: X-X.
We show that it is a birational isomorphism. For this purpose it is
sufficient to verify that
(1)
For PxCP= 1 on U, and by (1) f coincides on f-1(U) with the

isomorphism cp-l. The equation (1) is equivalent to the relation
(U x Y)nX = cp(U) , (2)
that is, to the fact that cp(U) is closed in U x Y. But this is obvious,
because cp(U) coincides in U x Y with the graph of the morphism
II1pk' Here f is an epimorphism because f(X») u, and U is dense in X.
It remains to show that X is projective. To do this we use the
projection g: X x Y - Y and show that its restriction g: X- Y is a
closed embedding. Since the notion of a closed embedding is local, it is
sufficient to find open sets V; C Ysuch that X C Ug-l(V;) and g determines
a closed embedding of X ng-1(V;) in V;. We set
V; = pi l(U i ),
where Pi: Y - Yi are the projections. First of all, the g-l (V;) cover X. To
see this it is enough to show that
(3)
because UUi=X and Uf- 1 (U i)=X. In its turn, (3) follows from the
fact that
(4)
It is sufficient to verify (4) on some open subset We f -1 (Ui ). In
particular, we may take W=f- 1 (U)=q>(U) [in accordance with (1)],
and then (4) is obvious.
Thus, it remains to verify that
g: X ng- 1 (Jti)_ Jti

determines a closed embedding. We recall that
Jti = Pi (Vi) =Ui x Yi, Yi =

-1 A A n lj,
j*i
-1
g (Jti) = X Ui X Yi.
A
We denote by Zi the graph of the morphism Ui x Y;-X, which is the

composition of the projection onto Ui and the embedding in X. The
set Zi is closed in X x Ui x Y;=g-I(Jti), and its projection onto
Ui x Y; = Jti is an isomorphism. On the other hand, q>(U) C Zio and since
Zi is closed, we see that X ng-I(Jti) is closed in Zi. Therefore, the
restriction of the projection to this set is a closed embedding, and
Chow's lemma is proved.
Similar arguments show the analogous fact for arbitrary varieties,
where X in this case is quasiprojective (see Exercise 7).
2. The a-Process Along a Subvariety. Chow's lemma shows that an

arbitrary variety is fairly close to a quasiprojective one. Nevertheless
these are distinct concepts. Simple examples of non-quasi projective
varieties will be constructed in § 2.3. This construction makes use of a
generalization of the a-process defined in Ch. II, § 4. The difference
consists in the fact that now we construct a morphism (J: X' - X for
which the rational morphism (J-I blows up not a point Xo E X but a
whole smooth subvariety. The construction follows very closely the
case we have already discussed.
a) The Local Construction. According to Theorem 5 of Ch. II, § 3,
for every closed point of a smooth subvariety Y of a smooth variety X
there exists a neighbourhood U and functions U 1 , ... , Urn E @x(U),
m = codimx Y, such that a y = (u l , ... , urn) in @x(U) and that dxUl' ... , dxUm
are linearly independent at any closed point x E U (the latter condition
means U 1 , ••• , Urn can be included in a system of local parameters). If
these conditions are satisfied, then we say that Ul' ... , Urn are local
parameters of the subvariety Yin U.
Suppose that X is affine and that Y has local parameters U 1 , ••• , U m

in the whole of X. We consider the product X x IP m- 1 , and in it the
closed subvariety X' defined by the equations tiuix ) = tjui(x),
i,j= 1, ... ,m, where (t 1 , ... ,tJ are homogeneous coordinates in ]pm-1.
The projection X x ]pm - 1 ~ X determines a morphism a: X' ~ X.
Here a- 1(y) = Y x ]pm -1, and a determines an isomorphism
X' -(Yx IPm-1)~X - Y.
If x' = (y, z) is a closed point on X', y E X, Z E IPm-I, Z = (Z1 : ... : zJ and

Zj :;60, then in a neighbourhood of x' we have uj=Ujs j, Sj= titj. Let
V1, ... , Vn - m ' U1' ... , Um be a system of local parameters at a point y
on X. Then the maximal ideal of the point x' on X' has the form
m x '=(v 1, ... ,Vn - m' U1, ... ,Um, S1- S1(X /), ""sm-sm(x /»)
=(v 1, ... , Vn - m ' S1 -S1(X /), ... ,A(x/), U;, ""sm-sm(x/»).
Hence, as in Ch. II, § 4.2, it follows that X' is smooth, n-dimensional,
and irreducible. Just as there, so we have here:
Lemma. If the a-process 't': X ~X is determined by another system of
parameters v1, ... , Vm of the same subvariety YCX, then there exists an
isomorphism cp : X' ~ X for which
X' cP IX
\/ X
is commutative. This isomorphism is unique.
On the open sets X' -a- 1(y) and X -'t'-1(y) we have cp='t'-1 a,
and the uniqueness follows from this. By definition, in these sets
cp(x; t1: ... : tm) = (x; V1(x): ... : vm(x») ,
1p(x; t~: ... : t;,.) = (x; U1(x): ... : um(x») ,
where 1p = cp - 1.
By hypothesis,
Vz= 'L,hZ,jUj , hZ,jEk[X]. (1)
j
In the open set tj :;60 we write Sj=titj, we express (1) in the form
Vz=Ujgz, gz= 'L,(a*hz,)sj' (2)
j
and we set
(3)
The same simple verification as in the proof of the analogous lemma

in Ch. II, § 4.2, shows that <p is a morphism and coincides with the
one already constructed on X' - (J - 1 (Y). The construction of 1p is
similar.
b) The Global Construction. Let X = UU" be an affine covering
such that Yis defined in U" by the equations U",l' ... , u",m' By applying
to U" and Yn U" the construction under a) we obtain a system of
varieties X~ and morphisms (J,,: X~ ~ U". The relation X~) (J; 1 (U" n Up)
holds for any IX and /3, and by the lemma there exist uniquely
determined isomorphisms
<P",p: (J; l(U"n Up)~(Jil(U"n Up).
It is easy to check that they satisfy the conditions for pasting
together and determine a variety X' and a morphism (J': X' ~ X.
The so constructed morphism is called the (J-process with centre in Y.
From the lemma it follows in an obvious way that neither X' nor
(J depend on the covering {U,,} or the system of parameters U",i'
c) The Exceptional Subvariety. The subvariety (J-1(y) is known
to us locally:
(4)
Globally we are concerned here with a fibering of a new type:
(J-1(y), y E Y, is a projective space. The relation (4) shows in what sense
our fibering is locally trivial.
With every vector bundle p: E~X we can connect a fibering
<p: lP(E)~X of this type. For this purpose we define lP(E) as the set
lP(E) = UlP(EJ,
xeX
where lP(Ex) is the projective space of lines of the vector space Ex.
To equip lP(E) with the structure of an algebraic variety we consider
a covering X = UU" in which E is given by transition matrices C".P'
Having fixed an isomorphism p -1 (UJ c:::. U" x V, where V is a vector
space, we obtain a mapping
U lP(Ex) ~ U" x lP(V) ,
xeUa;
by means of which we can introduce in this space the structure of an

algebraic variety. Obviously all these structures are compatible with
each other and determine on lP(E) a unique structure of an algebraic
variety. This is called the projectivization of the vector bundle E.
Specifically, lP(E) is pasted together from open sets
by means of a rule for pasting that is determined by the automorphisms

of the variety (U"n Up) x lP(V):
qJ",p(u,~) = (u, lP(C",p) ~), (5)
where u E U"n Up, ~ E lP(V), and IP(C",p) is the projective transforma-

tion with the matrix C",p'
We return to the variety O'- 1 (y) that arises in the O'-process
0': X' - X. It is pasted together from the open sets (Yn UJ x IPm-1,
and the rule of pasting is given by the formulae (1). This rule falls
exactly under the type (5) if for C",p we take the matrix
C",p = (hdy).
The functions hi,i are defined by (1), and a single glance at the
transition matrices of the normal bundle-formulae (9) in § 1.3-is
sufficient to convince us that the C",p correspond to the bundle N x/yo
Thus, we may express the result of our discussion by the simple
formula
O'- 1 (y) ~IP(N x/y),
d) The Behaviour of Subvarieties.
Proposition. Let Z be a closed irreducible smooth subvariety of X,
transversal to Y at each of their points of intersection, 0': X' - X the
O'-process with centre in Y. Then the subvariety O'- 1 (Z) consists of two
irreducible components:
O'- 1 (Z) = O'- 1 (YnZ)uZ' ,
and 0': Z'_Z determines the O'-process of the variety Z with centre in
YnZ.
The proof follows very closely the arguments in Ch. II, § 4.3.
Our problem is local, therefore we may take it that YnZ has in X
the local parameters U1, ••• , u" and that among them u 1 , .•• , ur , ••• , U m
are parameters for Y, and ur + 1, ... , Um' .•• , U l for Z. Then X' is
determined in X x IPm - 1 by the equations
(6)
We denote by Z the closure of the set O'- 1 (Z -(YnZ»). Evidently

O'- 1 (Z)=O'- 1 (YnZ)uZ. Since U r + 1 = ... =UI=O at every point of
O'- 1 (Z-(YnZ» and at least one of U1,""Ur #O, we have on Z
tr + 1 = ... = tm = 0 .
Therefore
ZCZ x IPr-1,
where t 1 , ... , t, are homogeneous coordinates in lP,-l and on Z the

following relations hold:
These relations determine the O'-process 0': Z' -+Z with centre in YnZ.
So we see that Z c Z', and since both varieties have one and the same
dimension and Z' is irreducible, we have Z = Z'.
The subvariety Z' C X' is called the proper inverse image of the
subvariety Z C X under the O'-process.
In conclusion we make a few remarks in connection with the
notion of au-process.
1. It can be shown that a O'-process does not lead us out of the
class of quasiprojective varieties. We do not give a proof here.
2. The existence of O'-processes whose centres are not points
creates a whole range of new difficulties in the theory of birational
isomorphisms of varieties of dimension greater than 2. In particular it is
not known to what degree the results we have obtained in Ch. IV, § 3.4
for surfaces can be carried over to them. It is only known that not
every morphism X -+ Y that is a birational isomorphism splits into a
product of O'-processes. A relevant example was constructed by
Hironaka. Whether it is true that every birational isomorphism is a
product of O'-processes and their inverse morphisms is unknown at
present. On the other hand, the theorem on the elimination of points of
inderminacy by means of O'-processes is true in any dimension if k is a
field of characteristic 0; this was also proved by Hironaka.
3. Example of a Non-Quasiprojective Variety. The variety we are

going to construct by way of example is complete. If a complete
variety were isomorphic to a quasiprojective variety, then by the theorem
on the closure of the image it would be projective. Consequently, it is
sufficient to construct an example of a complete but non-projective
variety.
The proof that it is non-projective is based on the fact that the
intersection indices on projective varieties have certain specific properties.
Therefore we begin with some general remarks on intersection indices.
We shall make use of concepts that are a very special case of the
ring of classes of cycles of which we have talked in Ch. IV, § 2.3.
In our special case the required definitions are easy to give in-
dependently. Let X by a three-dimensional smooth complete variety,
C an irreducible curve, and D a divisor on X. We assume that
C rt. Supp D. Then the restriction edD) defines a locally principal
divisor on C (we do not assume C to be smooth) for which the
intersection index is defined (see the remark in connection with the
definition of intersection index, Ch. IV, § 1.1). In this case the inter-
section index is denoted by deg (ldD) and is also called the intersection
index of the curve C and the divisor D:
(C, D) = deg (ldD).
The arguments of Ch. IV, § 1 show that this index as a function of D

is additive and invariant under equivalence. In particular, the index
(C, ,1) is dermed, where ,1 is a divisor class containing D. Besides,
in the application we need only the case when C is a smooth curve,
and then both these properties are obvious.
We consider the free Abelian group AI generated by all the curves
C C X. For an element a E AI the index (a, ,1), ,1 C Cl X, is defined by
additivity. On AI we introduce an equivalence relation: a ~ b if
(a, ,1) = (b, ,1) for any divisor class ,1. In that case a and b are called
numerically equivalent.
We consider an example, which is basic for what follows. If
a = Lni C;, a' = Lnj Cj, and if all the curves Ci , Cj lie on a smooth
surface Yc X and a'" a' as divisors on Y, then a ~ a'. Indeed, for every
divisor D on X the restriction operation (l~i(D) can be carried out in
two stages:
hence for a E Div Y

(a, Dh = (a, (l~D}y.
From this our assertion follows by virtue of the invariance of the inter-
section index on Y under equivalence of divisors.
The preceding arguments referred to any complete variety X. The
'*
projectiveness of a variety X implies an important property: if
a = Lni Ci , ni > 0, then a O. In fact, for the intersection of an
irreducible curve C with a hyperplane section H of X the formula
(C, H) = deg C
is obvious, in particular, (C, H) > O. Therefore also (a, H) = Lni( C;, H) > O.
Before proceeding to the construction of the example we consider
an auxiliary construction. Let C 1 and C z by two smooth curves in a
smooth three-dimensional variety V, where C 1 and Cz intersect
transversally at Xo' Suppose that the curves C1 and Cz are rational.
Although our results are true independently of this fact, this assumption
simplifies the deductions somewhat. We denote by a: V' -+ V the
a-process with centre in C 1 . According to the proposition in § 2.2
a -1 (C z ) consists of two components:
a- 1 (C z )=a- 1 (x o)uC; ,
where a: C~ ---+C 2 is the a-process at Xo E C 2 • Hence in our case it is an

isomorphism. We denote the surface a- 1(C 1) by S1' As a very simple
exercise on the formula defining the a-process we recommend the
reader to verify that S1 and C~ intersect in the single point xo,
a (x o) = x o, and transversally. We denote the fibre of the morphism
a:S 1---+C 1
at each point x E C 1 by kx . Since we have assumed that C 1 is a rational

curve, any two points on it are equivalent: x' ~ x", hence on S 1
kx,~kx" .
The situation in which we find ourselves is illustrated in Fig. 16.
surface S1
fibrek x
Fig. 16
We now consider the a-process of the variety V' with centre in C~:
a: V---+ V' .
The inverse image (a)-1(S1) of the surface S1 is irreducible: according
to the proposition in §2.2, (a)-1(S1) = (O')-1(XO)US'1, and a: S~ ~S1
is the a-process of the surface S1 with centre at xo: Hence it follows
that (O')-1(xo)CS'1. On the surface S'1 we have (0')-1 (kxo)=lul',
r
where T= (a) -1 (x o), and a: ---+kxo is an isom~rphism. For x # Xo the
fibre O'- 1(kx ) is irreducible. We denote it by l~. By the preceding we
have on S'1
(1)
We denote by S2 the surface (O')-1(C~). Like S1, it is stratified over

C~ into fibres ~, y E C~, and on S2
(2)
The surfaces S'1 , and S2 intersect on the line T. Their mutual

disposition is illustrated in Fig. 17.
surface S;
lineskx line [i
Fig. 17
Now let us go over to numerical equivalence. Substituting (1) in (2)

we find that
(3)
The basic feature of this relation is its asymmetry relative to
k" and P, which is connected with the order in which we have carried
out the O"-processes. We utilise this in the example, which we are now
about to construct.
We consider a smooth three-dimensional variety V and in it two
smooth rational curves C l and C2 , which intersect transversally in two
points Xo and Xl (for example, V)1P 2 , Cl and C2 are a line and a
conic in JP2). In the variety Vo = V-X 1 we carry out, as before,
O"-processes first in Cl - Xl' and then in the proper inverse image of the
curve C2 - Xl' So we obtain a morphism
0"0: Vo - V-Xl'
In Vl = V - Xo we carry out the O"-processes in the opposite order:
first with centre in C2 - Xo, and then in the proper inverse image of the
curve C l - Xo. So we obtain a morphism
0"1: Vl - V-xo·
Evidently, the varieties O"Ol(V_XO-Xl) and 0"1 1 (V-x o -x l ) are
isomorphic, and the morphisms 0"0 and 0"1 agree on them. For the
curve Cl U C2 - {xo, xtl is disconnected,therefore, both O"Ol(V- Xo - Xl)
and 0"1 1 (V-xo-x l ) can be obtained by carrying out in V-XO-Xl
the O"-process with centre in Cl - Xo - Xl in the open set V - C2 , and
the O"-process with centre in C2 - Xo - X 1 in the open set V - Cl , and
then pasting together the resulting varieties with respect to the set
V- (C l U C2 ), on which the two O"-processes agree.
Thus, we may paste the varieties Vo and Vl with respect to the open
subsets O"Ol(V_XO-Xl) and O"ll(V-xo-xl) and obtain a variety
Vand a morphism
0": V- V.
In V the relation (3) holds, which we have derived using the existence
of a common point Xo on the curves C l and C2 • Similarly, the
existence of the point Xl leads to the relation
(4)
where i' is an irreducible curve. Substituting one relation in the other
we find that
hence
(5)
To arrive at a contradiction to the assumption that Vis projective
it remains to show that it is complete. For any variety Z the
projection V x Z - Z can be split up into the composition of the
mappings (0",1): VX Z- Vx Z and the projection Vx Z-Z. Since Vis
projective, the image of a closed set under the second projection is
closed, and we need only show the analogous property for (0",1). We
know that V is the union of the two open sets V - Xo and V-Xl' and
since the notion of closure is of local character, it is sufficient to
verify that
(0", l): (0", l)-l((V- Xi) x Z)-(V- Xi) x Z, i =0,1,
carries closed sets into closed sets. On the sets V-Xi the morphism
0" coincides with the composition of O"-processes, and it remains to show
that for any O"-process 0" : U' - U and any Z the morphism
(0",1): U' x Z - U x Z
carries closed sets into closed sets. Again the local character of the
problem allows us to assume that 0" is given by the construction
under a) in § 2.2, that is, U' C U x ]pm-l and 0" is induced by the
projection U x ]pm-l_ U. But then our assertion follows from the fact
that a projective space is complete: Theorem 3 of Ch. I, § 5.
Thus, if X were quasiprojective, then it would be projective, but this
is impossible, because the relation (5) cannot hold in a projective
variety.
The foundation of the argument on which the example is con-
structed are, of course, the relations (3) and (4). They lead to (5),
which cannot hold on projective varieties. Perhaps these relations
lines kx
Fig. 18
become clearer if they are illustrated in a very primitive way (Fig. 18).
Here the splitting of the fibre kxo into two components is shown as the
composition of the segment kxo from two segments: rand P.
Remarks. 1. The dimension 3 in this example is not accidental.
It can be shown that a 2-dimensional smooth complete variety is
projective. On the other hand, there exist examples of complete, but
not projective, two-dimensional varieties with singular points.
2. In our examQle we consider an affine open subset U C V. If both
the curves P and 't in (5) were to intersect U, then we could find a
divisor D for which (p . D) > 0, (1' . D) > 0, which contradicts (5). For D
we could take the closure ota hyperplane section of that affine space
in which U lies. Thus, Pand 't lie "very far apart" in V: if an open affin~
subset contains at least one point of P, then it does not intersect 't.
4. Criteria for Projectiveness. In conclusion we give some criteria,

which characterize projective varieties among arbitrary complete
varieties. We do not state them in the greatest possible generality. In
particular, in the first two we assume the variety to be smooth. This
could be omitted, but it would require some additional explanations.
1. Criterion of Chevalley-Kleiman. A smooth complete variety is
projective if and only if any finite set of its points is contained in an
affine open subset.
Evidently, on a projective variety X there always exists a hyperplane
section H that does not contain a given finite set S, so that Sex - H,
and X - H is affine. Therefore one half of the criterion is obvious. In
the example of the non-projective variety we have constructed above
this criterion obviously does not hold (see Remark 2 after the
example).
2. Criterion of N akai- M oishezon. A smooth complete variety X is
projective if and only if on it there exists a divisor H such that for
every closed subvariety Y
Here Qy(H) denotes the restriction of H to Y.

For projective varieties H can be taken to be a hyperplane section.
In that case
Therefore the criterion evidently holds for projective varieties.

In formulating the last criterion we recall that on the projective
space ]pn there is defined a line bundle E C ]pn X V, where V is the
vector space whose lines are represented by points of]pn (Examples 1
and 2 of § 1.4). Here the projection ]pn x V~ V determines a morphism
E ~ V, which coincides with the a-process in V with centre at the
origin of coordinates. In this representation the only exceptional
subvariety is the null section of the bundle E. Let Xc]pn be a closed
subvariety. The bundle E' = Qx(E), the restriction of E to X, is a closed
subset of E, and the a-process a: E ~ V determines a morphism
a' : E' ~ V. From the completeness of a projective space it follows that
a carries closed set into closed sets. Therefore, V' = a' (E') is an affine
variety. Obviously, the only exceptional subvariety for a' is the null
section.
These arguments establish the "only if' part of the following
criterion.
3. Criterion of Grauert. A complete variety is projective if and only
if on it there exists a line bundle E and a morphism f: E ~ V onto
an affine variety V such that f is a birational isomorphism and its only
exceptional subvariety is the null section of the bundle E.
More briefly, the condition of Grauert's criterion can be stated as
contractibility of the null section of E to a point.
Exercises
1. Give a new proof of Theorem 1 in § 1, using Chow's lemma and a reduction to
Theorem 3 in Ch. II, § 3.
2. Show that if X is a complete variety and u: X' .... X a u-process, then X' is also
a complete variety.
3. Show that IP (E) :dP (E') if E is a vector bundle, and E' = E 0 L, where L is a
vector bundle of rank 1.
4. Let X be a smooth complete variety, dim X = 3, Yc X a smooth curve,

u:X'-+X a u-process with centre in Y, YoE Y, l=u- 1 (yo). Show that (l,u*D) =0,
where D is any divisor on X and u* its inverse image on X'.
5. Under the conditions of Exercise 4, let S = u- 1 (y). Show that (I, S) = - 1.
Hint: Consider a surface D passing through Ythat is smooth at Yo, and apply to it the
result of Exercise 4.
6. Show that for every smooth projective 3-dimensional variety there exists a non-
projective variety birationally isomorphic to it.
7. Show that for any irreducible variety X there exists a quasiprojective variety X
and an epimorphism f: X -+ X that is a birational isomorphism. There exists an
embedding XC!pn x X such that f is the restriction to X of the projection !pn x X --> X.
§ 3. Coherent Sheaves
1. Sheaves of Modules. In connection with vector bundles we have
come across sheaves of bundles over a sheaf of rings (f)x. Such
sheaves are a particularly convenient tool in the investigation of
algebraic varieties. One example will be in this section. We now begin
with some general properties of these sheaves.
Let us consider a very general situation: a ringed space, that is,
a topological space X on which a sheaf (f) of rings is given. Later we
shall analyse sheaves on X that are sheaves of modules over (f). We do
not say so explicitly and simply speak of sheaves of modules. It is clear
that every sheaf of Abelian groups over a topological space X can be
regarded as a sheaf of modules over a sheaf of rings (f), if we take for (f)
the sheaf of locally constant functions with values in 7l.
The definition of a homomorphismf : fF -+ f§ of sheaves of modules
was given in § 1.3. We recall that this is a system of homomorphisms
fu: fF(U) -+ f§ (U) of modules over (f)(U) satisfying certain conditions
of compatibility.
Example 1. Let X be a smooth algebraic variety over a field k,
(f)x the sheaf of regular functions, Q1 the sheaf of one-dimensional
reg\llar differential forms. By assigning to f E (f)x(U) the differential
df E Q 1 (U) we define a homomorphism of sheaves
d: (f)x-+ Q1 .
It is a homomorphism of sheaves of modules over the sheaf of locally

constant functions with values in k, but not over the sheaf (f)x.
Our next aim is to define the kernel and image of a homomorphism
of sheaves of modules. The first definition is perfectly obvious. Let
f: fF -+ f§ be a homomorphism of sheaves of modules. We set
Jf"(U) = Kerfu. From the definition of a homomorphism it follows
that for U C V we have Q~ Jf"(V) C Jf"(U). Therefore the system
{Jf"(U), Q~} determines a presheaf. A simple verification shows that it
is a sheaf of modules. By definition this is the kernel of the

homomorphism f
The kernel of a homomorphism is an example of a subsheaf of ff.
This is the name for a sheaf of modules ff' for which ff'(U) C ff(U) for
all open sets U C X, where the homomorphisms Q~,?, are the
restrictions of Q~ ? to the modules of ff'(V).
The matter i~ somewhat more complicated with the concept of the
image of a homomorphismf : ff ~ t§. The fact is that the (9(U)-modules
5(U) = Imfu, together with the homomorphisms Q~,~, determine a
presheaf, which, speaking generally, is not a sheaf.
Example 2. Let X be a topological space, for the time being quite
arbitrary, (9 the sheaf of locally constant functions with values in the
field of real numbers lR, ff and t§ sheaves of continuous functions
with values in lR and in the circle G = lR;Z respectively. A homomorphism
f : ff ~ t§ is given by the fact that for U C X and cP E ff(U)
fu(cp)(x) == cp(x) (mod Z), x E U.
For every element 1p E t§(U) and every point x E U there exists a

neighbourhood Yx ofx such that the set 1p(v,,) eGis not the whole of G.
Then there exists a set TC IR such that the projection p: lR ~ lR/Z = G
determines a homeomorphism p: T ~1p(v,,). The function cp = p-l1p
then belongs to ff(v,,), and fvJcp) = 1p. In other words, Imfvx = t§(v,,).
On the other hand, in general, 1m fu 0;6 t§ (U). For example, let
X = G and let ~ E ~(X) be the identity mapping. It is easy to see that
this cannot be "lifted" to lR, that is, there is no continuous function
cp: G ~ lR such that cp(x) == x(mod Z). As we have seen, there exists a
covering G = UVa such that cp(1. = Q~ as Elm fVa' Evidently these func-
tions are compatible on the intersections Van Vp. However, there is no
function cp E 1m fG for which Q~ a cp = cp Condition 2) in the definition
(1.'
of a sheaf (Ch. V, § 2.3) is not satisfied.

There is a natural way of defining the image of a homomorphism
f: ff ~ ~ of sheaves of modules. We defined a presheaf 5' by the
condition
5'(U) = fdff(U)), U eX.
The sheaf 5 associated with the presheaf 5' (Ch. V, § 2.4) is called the
image of the homomorphism f and is denoted by 1m f
Recalling the definition of the sheaf associated with a pre sheaf we
see that 1m f is a subsheaf of ~ and that (1m f) (U) consists of the
elements a E ~(U) such that every point x E U has a neighbourhood
Uxfor which
It is obvious that f determines a homomorphism

ff --+Imf.
From the definition it follows at once that a homomorphism
f : ff --+ f§ for which Ker f = 0 and Imf = f§ is an isomorphism.
A sequence of homomorphisms ff JJ. ff 2 ... ~ ff n + 1 is said to be
exact If 1m h = Ker h+ 1, i = 1, .. :, n. If the sequence O--+ff .4 f§ -4;/f-.O
is exact, then ff can be regarded as a subsheaf of f§. Consequently
(Imf)(U) = f(ff(U») ,
that is, in the construction of the sheaf 1m f the transition from the
presheaf to its associated sheaf is superfluous. Therefore the sequence
O--+ff(U)~ f§ (U)~;/f(U) (1)
is exact for every open set U.
Example 2 shows that the sequence
O--+ff(U)~ f§(U)~;/f(U)--+O,
is not exact, in general (for example, when U = X). This fact is the
reason why there is a non-trivial cohomology theory of sheaves.
For every subsheaf ff of a sheaf f§ we can construct a homomorphism
f : <§ --+;/f such that Ker f = ff, 1m f = ;/f. To construct it we set
;/f'(U) = f§(U)/ff(U)
and define the homomorphisms e~,Jf" as the result of the action of the
homomorphisms e~,'W on these factor groups. This defines the presheaf
;/f'. For ;/f we choose its associated sheaf.
It is easy to verify that for the stalks of these sheaves we have the
relation
Therefore, an element a E f§(U) determines elements ax E;/fx for all

points x E U. An obvious verification shows that all the elements {aA
give an element a' E ;/f (U), and f: a --+ a' is a homomorphism with the
required properties. The sheaf;/f is called the factor sheaf of <§ by ff.
Evidently the sequence O--+ff --+ f§ --+;/f --+0 is exact.
Example 3. Let X be an irreducible algebraic variety over a field k,
and :K* the sheaf of locally constant functions with values in the
multiplicative group of k(X). The sheaf (!)* is defined by the fact that
(!)*(U) is the group of invertible elements of the ring (!)(U). It easy to
verify that for the factor sheaf ~ = %* /(!)* the group '@(U) is
isomorphic to the group of locally principal divisors of the variety U.
In this example %* and (!)* are regarded as sheaves of Abelian groups.
Definition. The support of a sheaf ff is defined as the set X - W,

where W is the union of all open sets U C X for which ff(U) = O.
This set is closed and is denoted by Supp ff.
Proposition. If S is the support of a sheaf ff and if U, V, U C V, are
two open sets for which U nS = VnS, then the homomorphism
Q~: ff(V)-ff(U) is an isomorphism.
Let a E ff(V), Q~a = O. By definition of S, every point x E V,
X ¢ S, has a neighbourhood VX' which we may regard as contained in V,
such that
Q~,.(a)=O.
By hypothesis, for points XES such a neighbourhood is U. From the

definition of a sheaf it follows that a = O.
Let aEff(U). We consider a covering V=U~ in which Uo=U
and U"nS = I:} for 0( =F 0 (for example, U" for 0( =F 0 are sufficiently
small neighbourhoods of points x E V, X ¢ S).
We set ao = a, a" = 0 for 0( =F O. From the conditions of the
proposition it follows that
Qg:,.,uiaJ=Qg!,.,u/ap) .
Hence, according to the dermition of a sheaf, there exists an element
a' E ff(V) for which v ,
Qu,.a =a",
in particular, Q~(a')=a for 0(=0. This proves the proposition.
From the proposition it follows that if S is the support of the sheaf,
then the modules ff(U) for all sets U having a given intersection
with S are canonically isomorphic. Therefore we can define a sheaf
!J! on S, by setting
for open sets DeS.

Example 4. Let X be a scheme and YC X a closed subscheme. We
define a subsheaf f y of the structure sheaf (!)x by the condition
f y(U) = Oy if U is an affine open set, U = Spec A, and Oy C A is the
ideal of the subscheme Y n U. Obviously, if U does not intersect Y,
then" ylu = (!)xlu· Therefore the sheaf ff = (!)xl" y is 0 on these open
sets, that is, its support is contained in Y. The corresponding sheaf
!J! is the same as the structure sheaf (!)y of Y.
Note. Our definition of the support of a sheaf is not the one
generally accepted, however, it is somewhat more convenient for our
purposes. Besides, the two dermitions coincide in those situations in
which they are going to be applied later.
2. Coherent Sheaves. In discussing vector bundles we have already

come a~ross locally free sheaves. Now we consider a class of sheaves
that is in the same relation to an arbitrary module of finite type as
locally free sheaves are to free modules of finite rank.
The concepts introduced in § 3.1 are now going to be applied to the
case when (X, (l)x) is an arbitrary scheme. We begin with the local
analysis and assume that X = Spec A, where A is an arbitrary ring.
For every module M over A and every multiplicative system S of
elements of A we define the localization of M relative to S by setting
Ms=M®AAs·
The module Ms can be described in the same way as the localization
As of a ring in Ch. V, § 1.1: it consists of pairs (m, s), mE M, s E S,
with the same rules of identification, addition and multiplication by
elements of As as in the case of rings. We write the pair (m, s) in the
form mls. In particular, by taking for S the system of powers of a
non-nilpotent element f E A we obtain a module M f over the ring A f'
The homomorphisms As~As', which are defined for S C Sf, give
rise to homomorphisms Ms~Ms'. This enables us to associate with a
module M over a ring A a sheaf M on Spec A. Its defmition simply
copies that of the sheaf (I), into which it turns in case M = A. By
virtue of this we omit some verifications, which in the general case do
not differ at all from those made in Ch. V, § 2.2.
For an open set U = D(f),f E A, we set
For any open set U we consider all the f E A for which D(f) cU. For
them there are defined the homomorphisms
Mg~Mf
if D(g)) D(f). Using these homomorphisms we can define the projective

limit of the groups M f' We set
M(U)= JEt Mf ·
D(.f)C U
The group M(U) is a module over the ring (I)(U) = I~Af; this is a
general property of a projective limit. The inclusion U C V defines a
homomorphism e~:M(V)~M(U), just as in the case M =A. The system
(M(U), e~) defines a sheaf of modules M over the sheaf of rings (l)x.
Every homomorphism of A-modules cp: M ~ N determines homo-
morphisms cP f : M f ~ N f for all j E A.1 and after passage to the limit a
homomorphism of sheaves rp: M ~ N. If cp: M ~ Nand 1p: N ~ L are
two such homomorphisms, then

qJ1p= {p ·liJ .
The module M can be recovered from the sheaf M. In fact, there
is a generalization of the relation proved in Ch. V, § 2.2,
M(SpecA)=M,
which is proved word-for-word in the same way. Hence it follows that the
correspondence M --+ M between modules M and sheaves obtained
from them is one-to-one. Furthermore, a simple verification enables us
to deduce that the correspondence qJ --+ {p is a homomorphism of the
groups
Hom (M, N) '" Hom (M, IV) ,
where on the left-hand side we have the group of homomorphisms of
modules over A, and on the right-hand side that of sheaves of modules
over lDspecA-
Now we can go over to a globalization of these concepts.
Let X be a Noetherian scheme.
Definition. A sheaf :F on X is called coherent if every point
x E X has an affme neighbourhood U such that U = Spec A, A is a
Noetherian ring, and the sheaf :Flu is isomorphic to a sheaf of the form
M, where M is a module of finite type over A.
Proposition. If X is an affine Noetherian scheme and X = Spec A,
then every coherent sheaf:F on X is of the form M, where M is a module
offinite type over A.
Proof Since the open sets of the form D(f) determine a basis of the
spectral topology, there exist elements j; E A such that UD(j;) = X and
on D(j;) the sheaf :F is isomorphic to one of the form M;, where M;
is a module of finite type over the ring A Ii. The space Spec A being
compact, we may assume that the elements j; are finite in number. We
set :F(X) = M and show that :F = M.
For every non-nilpotent element g E A there is defined a homo-
morphism (2~(g): M --+:F(D(g)), which owing to the fact that :F(D(g)) is a
module over Ag can be extended uniquely to a homomorphism of Ag-
modules
qJg: M(D (g))--+:F (D(g)).
An obvious verification shows that this system of homomorphisms

determines a unique homomorphism of sheaves of modules M --+:F.
We show that qJ is an isomorphism.
Everything reduces to proving that the homomorphism qJg is an
isomorphism. To do this we consider the sequence of homomorphisms
O--+M ~ EBM;-4EBM;.j' (1)
where
M;,i = (M;)fj = (M)f, = ff(D(fdj») , A,(m) = (... Q~(f,)(m) ... ),
/l( ... m; ... mj ••• ) = ( ... (QZ~~:~j)(m;)-~~~{}im) ... ).
In the sequence (1) M; and M;,j are regarded as A-modules. From the
definition of a sheaf it follows that this sequence is exact. Now we use
the important but trivially verifiable property of the functor M ~Mg:
it carries exact sequences into exact sequences. In particular, the
sequence
O~Mg~ $(M;)/~ $ (M;)g
is exact. On the other hand, consider the sheaf ff ID(g)' It has a similar
exact sequence
O~ff(D(g»)~EBff (D(gji»!iEBff (D(gji./j») .
i i,j
But ff (D(gji») ~ (M;)g, ff (D(gji./j») ~ (M;)g. These isomorphisms induce

an isomorphism cp~:Mg~ff(D(g»). It is easy to check that CP~
coincides with CPg on the images of elements of M, and therefore on
the whole of Mg. So we have shown that cP is an isomorphism and
hence that ff =A1.
It remains to show that the module M is Noetherian when it is
known to us that the modules M;=M f , are Noetherian. Let {Mn} be
an increasing sequence of submodules of M. Then (Mn)f, = (M n+ l)f,
for all /; and sufficiently large n. Hence it follows that M n = M n + 1 •
3. Devissage of Coherent Sheaves. Here we give an account of

a method that enables us to reduce, although only in very coarse
problems, arbitrary coherent sheaves to free sheaves.
Proposition 1. For every coherent sheaf:F on a Noetherian irreducible
and reduced scheme X there exists an open dense set W such that the
sheaf fflw is free.
The assertion has local character, therefore we may restrict
ourselves to the case when X = Spec A, where A is a Noetherian ring
without nilpotent elements, and ff = it, where M is a module of
finite type over A. Furthermore, it is obvious that we may take X to
be irreducible. Under this assumption it follows that A has no divisors
of zero.
We recall that the rank of an A-module is the maximal number of
linearly independent elements over A. By hypothesis, the rank of M
is finite. We denote it by r and let Xl' ..• , Xr be linearly independent
elements over A. By definition, the submodule M' generated by them is
free. Let Yl' ... , Ym be a system of generators of M. Then there exist
elements d; #- 0, d; E A, such that

( 1)
We consider the open set W=D(d), d=d 1 ••. dm • The sheaf 3i'lw is
isomorphic to Md' But by (1) Md = M~, hence
3i'lw=M~.
The module M~ over Ad is free, because M' is free, and the proposition
is proved.
Proposition 2. For every coherent sheaf 3i' over an irreducible
Noetherian reduced scheme X there exists a coherent sheaf ';§ containing
a free subsheaf (!)r and a homomorphism cp : 3i' ~ ';§ such that the supports
of the sheaves Ker cp and ';§ j(!)r are different from X.
As will be clear from the proof, we can construct a homomorphism
cp : 3i' ~t§ for which the support not only of the sheaf Ker cp but
also of ';§lIm cp is different from the whole scheme X. Since the support
of ';§ j(!)r is different from X, Proposition 2 shows that every coherent
sheaf is free "to within sheaves of support other than X".
Proof Let W be the open set whose existence is established in
Proposition 1, and let f: 3i'lw~(!)rlw be the isomorphism that exists
according to the same proposition. We may assume that W is a
principal open set and will do this in what follows. We define the
sheaf ';§ by the condition:
,;§(U) = fUnweg nw 3i'(U) +egnW(!)r(u). (2)
Since egnW(!)r(u)C (!)r(u n W), and fUnweg nw3i'(U) C (!)r(u n W), both
terms on the right-hand side of (2) are contained in one and the same
group. We consider the sum of these subgroups, which obviously is a
submodule of (!)r(u n W) over the ring (!)(U), provided that for
x E (!)r(u n W), a E (!)(U), we set a' x = egnw(a) . x. Since the modules
3i'(U) and (!)r(u) are of finite type over (!)(U), so is the module ,;§(U)
over (!)(U).
The definition of the homomorphisms e~,rg is self-evident. From
what we have said above it follows at once that the sheaf ';§ we have
constructed is coherent.
The homomorphism egnw for (!)ris an embedding. It is sufficient to
verify this for an affine open set U = Spec A. We consider a principal
open set D(f) C Un W. The kernel of the homomorphism e~(f) consists of
those elements XEA such thatrx=O for some n;;;.O. Since the scheme X
is irreducible, the ring A has no divisors of 0, hence x = 0. A fortiori
Ker egnw = 0. Thus, we may identify (!)r with a subsheaf of ';§ by
means of the homomorphism eg(")w'
We define the homomorphism ({J:!IF -+ "§ by the conditions

({Ju = fUnwggnw .
If Uc W, then
"§ (U) = fuggnw!IF(U) = fu!IF(U) = (9'(U) = ggnw{9'(U)
and fu is an isomorphism. Therefore ({Ju is an isomorphism, and
"§ (U) = (9'(U). This shows that the sheaves Ker ({J and "§ /{9' vanish
on W, hence the supports of these sheaves are contained in X - W.
Proposition 2 leads to the question of the structure of coherent
sheaves whose supports differ from the whole sheaf. If the support of a
sheaf !IF is a closed set YC X, then according to § 3.1 on Ya sheaf
ff is defined by the condition
#(tJ)=!IF(U) if UnY=tJ.
We regard Yas a reduced closed subscheme of X. Is perhaps
!IF a coherent sheaf over this scheme or at least over the sheaf of
(9-modules on it? Generally speaking, this is not true, as the following
example shows. Let X = Spec'll, and suppose that !IF corresponds to
the module 'll/p 2'll, where p is a prime number. The support of !IF is the
ideal (P), and the reduced scheme corresponding to it is Spec (ll/p'll).
Obviously it is impossible to define on 'll/p 2'll the structure of a module
over the ring 'll/p'll.
However, we can show that in a weaker form the sheaf ff can be
made into a coherent sheaf over Y.
Proposition 3. If !IF is a coherent sheaf on a Noetherian scheme X,
with support Y:;6 X, then in the sheaf ff on the reduced subscheme Y
there exists a sequence of subsheaves ff = ff0 ) ff1 ) ... ) ffm = 0 whose
factors ff i/ffi + 1 are all coherent {9y-modules.
Proof In § 3.2 we have given an example of a sheaf Jy connected
with the subscheme Y. Clearly, the sheaf ff is coherent if the following
relation holds:
ff .J y = 0. (3)
For in this case all the (9x(U)-modules !IF(U) are modules over
(9x(U)/J y(U) = (9y(U). Therefore, if in the affine open set U = Spec A
the sheaf !IF is of the form M, where M an A-module, then M· Oy = 0,
hence M is a A/oy-module. Here ff = M if we now regard M as a
module over A/oy.
Let us show that a somewhat weaker assertion is true: there exists
a number m > 0 for which
(4)
We consider the affine open set U = Spec A, for which the

restriction of fJ' to it is of the form it, where M is a module of
finite type over A. Let uy C A be the ideal of the subset Yn U. Iff E uy ,
then D(f) C U - (U n Y), and by hypothesis the restriction of fJ' to
D (f) is zero. This means that M f = 0, hence for every mE M there
exists a j(m) > 0 such that f j(m)m = O. Since M is a module of finite type,
it follows from this that f jM = 0 for some j> O. From the fact that
this relation holds for every f E uy and that the ideal Uy has a finite
basis we deduce that
u~M=O (5)
for some 1> O. In other words, (4) holds for the open set U. Choosing
a finite covering of X by such open sets and taking for m the maximum
of those numbers I for which (5) holds on each of these factors, we
obtain (4) on the whole of X.
We set fJ'j = fJ' f~, i = 0, ... , m, fJ' 0 = fJ'. Obviously the supports
of all the sheaves fJ'j are contained in Y. We denote by #j the
sheaves determined by the fJ'j on Y. By (4) # m = fJ'm = O. Since
the relation (3) holds for the sheaves #j/#j + l' hence they are
coherent. This proves Proposition 3.
In conclusion we show that by the method we have used all the
time we can reduce the study of sheaves to the case of irreducible
schemes.
Proposition 4. Let X be a Noetherian reduced scheme, X = UXj a
representation in the form of a union of irreducible components, and ff' a
coherent sheaf on X. There exist coherent sheaves fJ'j on X and a
homomorphism ({J: fJ' -+ EBfJ'j such that the support of fJ'j is contained
in Xj, the sheaf #j determined on X j by fJ'j is coherent, and the support
of the kernel of ({J is contained in U (XjnXJ
j*j
Proof We set fJ'j=fJ'/fJ'fx" let <.pj:fJ'-+fJ'j be the natural
projection, and ({J = EB ({Jj. We have seen in § 3.2 that the support of
fJ'j is contained in Xj, and since fJ'jfx i = 0, the sheaf #j is coherent.
We consider the open set
Uj=X i - U(XjnX).
i*j
On it f Xj = @x for j #- i, and" Xi = 0, therefore fJ')Ui = 0 for j #- i,

and fJ';lUi = fJ'\Ui· Thus, <.pj = 0 for j #- i, and ({J = ({Jj is an isomorphism.
Therefore the kernel of ({J is 0 on UUj, and this is what we have
claimed.
4. The Finiteness Theorem

Theorem. If X is a complete variety over a field k and :F a coherent
sheaf on X, then the vector space :F(X) is finite-dimensional over k.
The basis of the proof is the following remark. If we are given a
homomorphism of sheaves over k
({J::F ~r:§ , £ = Ker ({J, dim £(X) < ex) , dim r:§(X) < ex) , (1)
then the space §(X) is finite-dimensional. This follows from the

definition of the kernel, according to which £(X) is the kernel of the
homomorphism ({Jx: :F(X)~r:§(X). Hence we obtain by induction
that the space :F(X) is finite-dimensional if there exist subsheaves
(2)
for which the spaces :Fd:Fi+ 1 (X) are finite-dimensional.

We prove the theorem by induction on the dimension of the
variety X. If dim X = 0, then X consists of finitely many points, the
coherent sheaf :F on X is, by definition, a finite-dimensional vector
space over k, and the theorem is proved.
We now assume that the theorem is true for complete varieties
whose dimension is less than that of X. We show that we can then
derive the theorem for all sheaves :F on X whose support is contained
in a closed subvariety Y, dim Y < dim X.
Indeed, by the definition of ff we have :F(X) = #"(Y), and we may
apply the theorem to the coherent sheaves on Y. Here we are faced
with the difficulty that, in general, the sheaf #" is not coherent on Y;
but the position is saved by Proposition 3 of § 3.3. It gives us a
sequence ff = #"0 ) #"1 ) ... ) ffm = 0, in which the sheaves ffi/ ffi + 1
are coherent on Y, so that the inductive hypothesis is applicable to
them. We obtain a sequence of sheaves (2), and from its existence
it follows that the space #"(Y), hence also :F(X), is finite-dimensional.
The next step of the proof consists in the reduction of the theorem
to the case of irreducible varieties. Let X = UXi be the decomposition
into irreducible components. Here we can apply Proposition 4 of
§ 3.3. Our homomorphism ({J has a kernel whose support is contained
in the subvariety U(XinX), which is of smaller dimension than X.
i*j
Therefore it is enough for us to show that the space (E8:F i) (X) is
finite-dimensional. But
(E8:FJ (X) = E8(#"i(XJ),
and since #"i is a coherent sheaf on Xi, this reduces the assertion to the
case of irreducible varieties Xi.
Finally, we come to the central part of the proof, assuming X to be

irreducible. Here we rely on Proposition 2 of § 3.3. Since X is
complete, we have lP(X) = k, hence dim lPr(x) = r. Since the support
of f§ /lP r is different from X, the theorem is true for this sheaf, hence
we have for f§ the homomorphism 1p: f§ ~ f§ /lP r which satisfies (1).
Therefore the space f§ (X) is finite-dimensional. On the other hand,
the homomorphism q>: ~ ~ f§ constructed in Proposition 2 of § 3.3
again satisfies (1), hence the space ~(X) is also fmite-dimensional,
as required.
The theorem we have proved has many important applications.
Some of these were mentioned earlier. First of all, in § 1.4 we have
associated with every divisor D on a variety X a sheaf .If'D such that
.If'D(X) is isomorphic to the space .If'(D) introduced in Ch. III, § 1.5.
We have seen in § 3.4 that the sheaf .If'D is locally free of rank 1, hence
coherent. Thus, our theorem is applicable to it, and we obtain the
result, which has already been used several times:
Corollary 1. The dimension I(D) of a locally principal divisor D
over a complete variety is finite.
Applying the theorem to the sheaf corresponding to the cotangent
bundle Q1 and its exterior powers QP we qbtain:
Corollary 2. On a complete smooth variety the dimensions hP of the
spaces QP[X] of regular differential forms are finite.
This result was also stated in Ch. III, where we have seen that it
gives a number of birational invariants of a variety.
As a further example we consider the sheaf f/ corresponding to the
tangent bundle. The elements of the group f/(X) are called regular
vector fields on X. Such an element can be regarded as a function
associating with every point x E X the tangent vector tx E ex at that
point. In this case our theorem yields:
Corollary 3. The dimension of the space of regular vector fields on a
smooth complete variety is finite.
Exercises
1. A coherent sheaf $i' is called a torsion sheaf if $i'(U) is a torsion module over (!)x(U)
for every open set U. Show that $i' is a torsion sheaf if and only if its support is
different from X (the scheme X is assumed to be irreducible).
2. Find the general form of torsion sheaves on a smooth curve.
3. Let E -> X be a vector bundle over an affine variety X = Spec A. Show that the
set ME of sections of E is a module of finite type over A.
4. Show that the module ME introduced in Exercise 3 is projective over the ring A
(for the definition of a projective module see [10], Ch. I, § 2).
5. Show that the modules ME and ME' are isomorphic if and only if E and E' are.
6. Show that every vector bundle over an affine line IA1 is trivial.
7. Let E-+X be a vector bundle over a complete vanety X. Show that the set of
sections ME is a finite-dimensional vector space.
8. Show that the set of morphisms f: El -+ E2 of vector bundles Ej-+X, j = 1, 2, over
a complete variety X forms a finite-dimensional space.
9. Let A be a one-dimensional regular local ring, % its field of fractions,
X = Spec A, x E X a generic point, U = {x}.
A sheaf ff of (!i-modules over X is given by an A-module M, a linear space L
over %, and an A-homomorphism <p: M -+ L. Express in these terms the' fact that ff is a
coherent sheaf. Construct an example of a subsheaf of a coherent sheaf that is not coherent.
to. Let X be an irreducible variety, Xo E X a closed point. We define a presheaf
ff on X by setting ff(U) = (!i(U) if U does not contain xo, ff(U)=O if U contains Xo.
Show that ff is a sheaf, that it is not coherent, and- that it is a subsheaf of (!i.
Part m. Algebraic Varieties over the Field
of Complex Numbers and Complex Analytic
Manifolds
Chapter VII. Topology of Algebraic Varieties
§ 1. The Complex Topology

1. Defmitions. In Ch. II, § 2.3, we have seen that the set of complex
points of an algebraic variety dermed over the field of complex numbers
is a topological space. In Ch. II this was shown for quasiprojective
varieties, the only ones at our disposal at that moment. But these argu-
ments remain valid for arbitrary varieties. Presently we shall give a
general definition. The topology that determines on X a given structure
of a scheme is called the spectral topology.
First we introduce some notation. For a variety X defined over the
field «:: of complex numbers we denote by X(<<::) the set of its closed
points. We consider a set U C X that is open in the spectral topology,
finitely many functions 11' ... ' 1m that are regular on U, and a number
e > O. We denote by V(U; 11> ... ' 1m; e) the set of those points x E U (<C) for
which
JIi(x)J <e, i= 1, ... , m.
We tum the set X(<C) into a topological space, by taking the sets
V(U; 11' ... ' f,.;e) as a basis of open sets.
The topology so dermed is called the complex topology. Let us
compare it with the spectral topology, which we have considered earlier.
If Y C X is a closed subset in the spectral topology, then Y(<<::) c X(<<::).
From the definition it follows that Y(<<::) is closed in X(<<::) in the complex
topology and that the complex topology of the set Y(<<::) is the same as its
topology as a subset of X(<<::). However, not every set that is closed in the
complex topology is of the form Y(<<::), where Y is closed in X in the
spectral topology. An example is the set of points x E A 1 (<<::) for which
Jt(x)J..;;; 1, where t is a coordinate on Al. The morphism I:X -+ Y of
algebraic varieties evidently determines a continuous mapping
I :X(<<::)-+ Y(<C).
In some respects the complex topology is simpler than the spectral
topology. As a very simple example we show that (Xl x X 2 ) (<<::) in the
310 Chapter VII. Topology of Algebraic Varieties
complex topology is the product of Xl(CC) and X2 (CC). It is clear that
V(Ul;fl' .·.,fm;e) X V(U2 ;gl' ... ,gn;e)

= V(Ul x U2 ;pt fl' ···,ptfm,pj,gl' ···,pj,gm;e),
where Pl and P2 are the projections of Xl x X2 onto Xl and X 2 • Therefore

products of open sets in Xl (CC) and X2 (CC) are open in (Xl x X2 )(CC). To
verify that they form a basis of open sets it is sufficient to do this for
affine Xl and X 2 • Embedding them in affine spaces we reduce the verifica-
tion to the case Xl =/An" X2 =/An 2 , where it is obvious.
In the complex topology X(CC) is a Hausdorff space. For by the
definition of a variety the diagonal L1 is closed in X x X in the spectral
topology. Therefore L1 (CC) is closed in (X x X) (CC) in the complex topology.
As we have just seen, (X x X)(CC) = X(CC) x X(CC), and L1(CC) coincides
with the diagonal of this space: the set of points of the form (x, x), XEX(CC).
The fact that the diagonal is closed is equivalent to the space being
Hausdorff ([8], § 8.1).
The topological space IPn(CC) is compact, hence so are all its closed
subsets. In particular, this refers to the spaces X(CC), where X is a pro-
jective variety. If X is a complete variety, then by using Chow's lemma
in Ch. VI, § 2.1, we construct a morphism f : X' -+ X, where X' is a
projective variety. This morphism is birational, hence f(X') is dense
in X, and since X' is projective,f(X') = X. In particular,f(X'(CC») = X(CC).
Since f is a continuous mapping and X'(CC) is compact, it follows that
X(CC) is compact. It can be shown that this property characterizes
complete varieties over the field of complex numbers: if the space X(CC)
is compact, then the variety X is complete (see Exercises 1 and 2 to
Ch. VII, § 2). It is evident that for an arbitrary variety X the space X(CC) is
locally compact.
The arguments in Ch. II, § 2.3, can now be applied to the investigation
of the complex topology of any (not only quasiprojective) smooth variety
defined over the field of complex numbers. They show that in this case
X(CC) in the complex topology is a topological manifold of dimension
2 dim X.
The preceding definition admits the following generalization. We
consider an arbitrary field k and denote by k' its algebraic closure. Let X
be a scheme over k such that the scheme X x k Speck' is an algebraic
variety over k'. This scheme is called an algebraic variety defined over k.
An example is an affine or a projective variety over /( in which an ideal
has a basis consisting of polynomials with coefficients in k.
If X is an algebraic variety over k, then X(k) denotes the set of those
closed points x E X for which k(x) = k.
If k is the field IR of real numbers or the field G2 p of p-adic numbers,

then we can define in the set of points X(k) a topology in exactly the same
way as we have done this in the case k = <C.
If k = IR and X is a smooth variety, then X(IR) is a topological manifold
of dimension dimX. Henceforth we only consider the topological space
X(<C), except in § 4, where we study the space X(IR) in the case when X
is a curve.
We always consider the space X(<C) with the complex topology. For
the remainder of this section we consider a space X(<C), where X is
smooth. We use here a somewhat larger topological apparatus than in
the remaining parts of the book. But on the other hand, the results we
obtain are nowhere used later.
2: Algebraic Varieties as Differentiable Manifolds. Orientation. Let X

be an n-dimensional smooth variety over the field <C of complex numbers,
x E X a point (from now on we only consider closed points) and t 1 , ... , tn
a system of local parameters in it. As was shown in Ch. II, § 2.3, there
exists a neighbourhood U of x at X(<C) which by means of the functions
t 1 , ••• , tn is mapped homeomorphically onto a domain of the space <cn. In
view of this, every function on U can be regarded as a function of the
variables t 1 (y), ... ,tn(y), or of the real variables U 1 ' ... 'U m Vl, ... ,Vn>
iftj=uj+ivj.
Defmition. A real-valued function on U belongs to the class CO if it
is infinitely differentiable as a function of U 1 , .•• , Un; VI' ••• , Vno
As was shown in Ch. II, § 2.3, other local parameters t~, ... , t~ are
analytic functions of t 1 , ... , tno Therefore the definition we have given
does not depend on the choice of local parameters, and the concept is
well-defined.
It is easy to verify that our definition introduces in the topological
space X(<C) the structure of a differentiable manifold ([26J, §1).
There is a natural connection between the properties of the algebraic
variety X and the differentiable manifold X (<C). Differential forms
OJ E QP [XJ are (complex-valued) differential forms on X (<C). If E ~ X is a
vector bundle, then E(<C)~ X(<C) is a topological vector bundle. Here
we only have to ignore that Ex is a vector space over <C and have to
regard it as a space (of double the dimension) over IR. In this correspond-
e
ence a tangent bundle ~ X corresponds to a tangent bundle of the
differentiable manifold X(<C).
Presently we consider the question of orientation of the differentiable
manifold X (<C). First we recall the relevant definitions.
An orientation of the one-dimensional vector space IR is one of the two
connected components of the set IR - 0; an orientation of an n-dimensional
vector space F is an orientation of the one-dimensional space An F.
An orientation of a (locally trivial) vector bundle f : E ~ X is a col-

lection of orientations COx of the fibres Ex such that every point has
a neighbourhood U and an isomorphism f -l(U)~ U x F carrying all
the orientations COx' X E U, into one and the same orientation of F. An
orientation of an differentiable manifold is an orientation of its tangent
bundle.
Proposition. If X is a smooth variety over <C, then the differentiable
manifold X(<C) is orientable.
The reason for this is very simple: if an n-dimensional vector space F
over <C is regarded as a 2n-dimensional space over lR, then it has some
canonical orientation. To define it we choose a basis e 1, ... , en in F
over <C. Then the vectors U1, ... ,U2n=e1,ie1, ... ,emien form a basis
of F over IR and have the orientation U 1 1\ ... II. U 2n of this space. Let us
verify that this orientation does not depend on the choice of basis
e1, ... , en' Let flo ... In be another basis of F over <C. We denote by cp
the linear transformation over <C that carries e 1, ... , en into f1, .. . ,fm
and by cp the same transformation cp in F, regarded as a linear space
over IR. We have to show that detcp > 0, and this follows from the identity
detcp = [detcp[2 . (1)
To prove it we consider the space F = F@ IR <C, and in it the transforma-
tion I defined by
I{f@rt)=if@rt,fEF,rtE<C. (2)
Then F = F1 EB F2, where F1 and F2 are the eigenspaces for I corresponding
to the eigenvalues i and - i. By analogy with (2) we extend cp to F. Of
course, its determinant remains unchanged. It is easy to see that F1 and F2
are invariant under cp and that the matrix of cp in F1 is the same as the
matrix of cp in F, and in F 2 is its complex conjugate. Equation (1) follows
from this.
Now let f:e~X(<C) be the tangent bundle and COx the canonical
orientation in ex,XEX. We show that by this the orientation in X is
determined. If U C X is such that
(3)
is an isomorphism of algebraic vector bundles, then a fortiori this is true
for the corresponding differentiable bundles. But in (3)
tpx: ex~F
is an isomorphism of complex vector spaces. Therefore it carries the
canonical orientation COx in e into the canonical orientation co in F. This
From now on the orientation we have just constructed is called the
canonical orientation of the manifold X (<C).
We have obtained a first restriction, which shows that not every

even-dimensional manifold can be represented in theform X(<C), where X
is some smooth algebraic variety. For example, such a representation
does not exist for the real projective plane.
3. The Homology of Smooth Projective Varieties. The orientability of

a differentiable manifold can be expressed in terms ofits homology. We
recall this connection (see, for example, [19], 251). The orientation W
of an n-dimensional vector space E over 1R determines an element of the
relative homology group W E Hn(E, E - 0, Z). If U is a chart containing
°
the point x ofthe manifold M, and qJ : U ~ E a diffeomorphism carrying U
into a neighbourhood of in E, then we have the excision isomorphisms
Hn(U, U -x,Z)~Hn(M,M -x,Z),

Hn(qJ(U), qJ(U) -0, Z)~Hn(E, E -O,Z)
and the isomorphism
Finally, dxqJ is an isomorphism of the tangent spaces
dxqJ:ex~E .
Using this system of isomorphisms we can associate with the orientation

Wx of the tangent space ex a homology class for which we preserve the
same notation:
Wx E Hn(M, M - x, Z) .
The orientation of the compact manifold M then determines a class
WM E Hn(M, Z), which is uniquely characterized by the fact that under the
homomorphism
corresponding to any point x EMit is mapped into Wx. The class WM is

called the orienting class of the manifold M.
The proposition in § 1.2 shows that if X is a smooth complete algebraic
variety, then X(<C) has a canonically dermed orienting class
WX(C[) E H 2n (X(<C), Z), n = dim X. In what follows we talk occasionally of X
itself as of a 2n-dimensional homology class, meaning by this the class

WX(c[)·
The preceding arguments construct the class WX(C[) E H2n (X(<C), Z),
which is necessarily non-zero, because it determines a non-zero class in
the group H 2n (X(<C), X(<C) - x, Z). For the same reason this class is of
infinite order. Hence for a smooth complete variety X
This is a special case of the following more general result.

Proposition. For a smooth projective variety X of dimension n
We can indicate 2/-dimensional cycles on X(CC) that are not homolo-

gous to O. For this purpose we consider a smooth subvariety Y C X of
dimension I, for example, the corresponding section by a linear subspace
of a projective space lPN containing X. Let j denote the morphism of
embedding Y in X and also the embedding Y(CC)-X(CC). The homology
class we consider isj*OJ y • We show that it is not homologous to O. Some-
what inaccurately, but more intuitively, we can express this by saying
that smooth subvarieties are not homologous to 0 in an enveloping
variety.
First of all, we can make a simple reduction. By combining the
embedding Y - X with the embedding X _1P N we see that it is sufficient
to prove our proposition for the composite embedding Y _ lPN. In other
words, we may assume that X = lPN, N = n.
The only result we use is Stokes' formula ([26], 34), from which it
follows that if <p is a closed m-dimensional differential form on some
manifold, OJ an n-dimensional cycle, and
J<p #0,
w
then OJ is not homologous to O.

The form we construct is defined over the manifold 1PN(CC). To
write it down we make use of the fact that a differential form over CCn can
be expressed in terms of the differentials dz 1 , ••• ,dzm dz 1 , ••• ,dzn ([9],
171). Consequently such an expression is also possible in local coordinates
on an algebraic variety. If <p is the form, we denote by d' <p its partial
differential with respect to the variables Z l' ... , Zn only, and by d" <p with
respect to the variables Zl, ... , zn. Then d<p = d' <p + d" <po
e
Let (0, ... , (n be homogeneous coordinates in 1Pn and a linear form.
e
In the domain # 0 we set
aa
2
<p=l'd'd"H =1L..
." H
a dZk /\ d-Zj'
I'-t12
k,j Zk Zj
H=log,~o
n
Replacing ~ by another form 11 does not change cp in the domain where

~ # 11 # 0, because
~
d'd" log 1 12 = °.
Thus, the form cp does not depend on the choice of ~, and by considering
various ~ we can define cp as a form on the whole of ]pn(Q::). This form is
closed owing to the fact that d d' d" = (d,)2 d" + d' (d")2 = 0.
Let us show that if Y is a smooth I-dimensional subvariety of]pn, then
(1)
roy
where cpl is the I-th exterior power of cpo This then gives us the required
result. The inequality (1) is a consequence of the following sharper
assertion.
Let t 1, ... , tl be local parameters in a neighbourhood of some point
Y E Y, tj = u j + ivj,j = 1, ... , I. We treat U 1 , V 1 , ... , UI, VI as local parameters
over the differentiable manifold Y(Q::), and we letj: Y(Q::)~]pn(Q::) be the
embedding. Then the form j* cp I on Y has in these local parameters the
form
where F is a real function and

F>O. (2)
Compared with (1) the inequality (2) has the advantage that it bears
purely local character. We now proceed to verify it.
The property of cp on which the proof of the inequality (2) is based is
the following. We write cp in local coordinates, and assuming, for example,
that Co # 0, we set Zi = (;/Co' Then
and here (c1J is the matrix of a Hermitian positive-definite form.

The property of being Hermitian is evident from the fact that
Positive-definiteness means that

I
l,j
CI,j~l~j > ° (3)
for any ~1' ... , ~n not vanishing simultaneously. The left-hand side of
(3) can be rewritten in the form
d~dpog( 1+ itl IZiI2),

where d~ is the differential with respect to the variables Z l ' ... , Zn in the
direction ~ = (~1' ... , ~n)' that is,
d~f=I uZi
~f ~i'
and similarly
U sing this, the left-hand side of (3) can be rewritten in the form
(1 + ~Izlf ((II~iI2)(1 + IlziI 2)-II zieiI 2),

and by Cauchy's inequality this is not less than
II~iI2
(1 + LlZil2f .
The remaining computation is simple linear algebra. If t 1, ... , tl are
local parameters on Y, then by substituting the expression
dz.
dz.= , \ , _ J dt l
J L... dt l
in the expression for cp we see that

I
j* cp = i I g«,pdt« A dtp .
«,P=l
Here g«,p is obtained from Cl,j by means of the formula for the restriction
of a bilinear form from the complex linear space 8y,lPn to the complex
linear subspace 8y,y. Therefore (g«,p) is also the matrix of a positive
definite Hermitian form. Finally, a simple calculation shows that
j* cpl = i l det(g«p) dtl A dtl A... A dt l A ifz
and if tj = Uj + iVj, then dtj A i0 = - 2iduj A dVj. Therefore in (2)
F = 2I det(g«p),
and the inequality (2) follows from the fact that the determinant of a
positive definite Hermitian form is a positive real number. This com-
pletes the proof of the proposition.
We give a sketch of another proof of this proposition. Although it

uses more topological tools, its idea is very lucid. We recall that if M is
a compact n-dimensional oriented manifold, then HP(M, <C) ® Hn- P(M,<C)
-+ Hn(M, <C) defines a duality between the spaces HP(M,<C) and Hn- P(M,<C).
Since HP(M, <C) is dual to Hp(M, <C), the spaces Hp(M, <C) and Hn- p(M, <C)
are also dual to each other. The corresponding scalar product is called the
intersection index or the Kronecker index of the two cycles. Let V and W
be two smooth oriented subvarieties of dimension p and n - p in M,
intersecting transversally. This means that the intersection consists of
finitely many points and in each such point x E V n W we have ex,M
= ex,vEB ex,w. In this case we can look at the embeddings jv: V -+ M,
jw: W -+M and exhibit for the intersection index Uv*w v -jw*ww) the
simple formula
Uv*wv'jw*ww)= L:
XEVnW
c(V, W,x), (4)
where c(V, W, x) is equal to + 1 or - 1 according as the natural orienta-

tion of e x,vEB ex, w is the same as the orientation of e x,M or opposite.
(For these results see [13], Ch. II.7.)
Now let M = X(<C), V = Y(<C), W = Z(<C), where X, Y and Z are
smooth complete algebraic varieties, YCX, Z C X, and Yand Z intersect
in X transversally. Then all the terms c in (4) are + 1. For in this case
ex,x, eX,y, and ex,z are complex linear spaces. The transition from a
complex basis of ex,yEB ex,z to a complex basis ex,x is effected by a
complex linear transformation. The corresponding real transformation
carrying a real basis of eX,yEBex.z into a real basis of ex,x has positive
determinant, as we have seen in § 1.2. Therefore
c(Y(<C), Z(<C), x) = + 1.
If X is a smooth projective variety, Y C X, and Y is smooth, then there
exists a smooth subvariety Z intersecting Y transversally with respect
to a non-empty subset. For Z we can take the section of X by the cor-
responding linear subspace. In that case (4) shows that
Uy*w y . jz*w z ) = deg Y
is equal to the degree of Y in the enveloping projective space. From this
it follows, of course, that jy* Wy i' O.
We can establish a connection between these arguments and the
earlier proof of the proposition. Namely, if X = lPn, and if Z is a linear
subspace of dimension n - I, then the homology class jz. Wz determines
by means of the intersection index a linear form on H 21 (X(<C), <C). It can
be shown that the form cp we have constructed corresponds, to within a
factor, to the cohomology class determined by a hyperplane section,
and <p' to the class determined by Z. More accurately, the class of hyper-
plane sections determines the form 21n <p. This means that the degree
of a smooth I-dimensional subvariety Y C IP" can be expressed as an
integral over the oriented cycle (Oy:
deg Y
1
= -(2 f I
)' . <p . (5)
n roy
In conclusion we remark that the proposition is probably true for
arbitrary complete varieties. However, the author does not know how
to prove it.
Exercises
1. Show that the formula (5) above remains valid when Y is a curve, possibly non-
smooth. Hint: Consider the normalization morphism.
2. Show that if X is a smooth projective curve, w e QI [X], and .r w = 0 for all
a
x
u e HI (X, 7L), then w = O. Hint: Consider the function <p(x) = .r w, where Xo is a fixed
and x an arbitrary point on X; show that the integral does not depend on the path of
integration, that the function <p is continuous and as a function of the local parameter at x
holomorphic. Show that the existence of a maximum of \<p(x)\leads to a contradiction to
the maximum modulus principle for analytic functions.
3. Show that the assertion of Exercise 2 is true for a smooth projective variety of
arbitrary dimension.
4. Show that if G is an algebraic group acting on a projective smooth variety X and if
the space G(CC) is connected, then g*w=w for geG, weQI[X]. Hint: Show that the
cycles u and g*u are homologous.
§ 2. Connectedness
The object of this section is to show that if X is an irreducible algebraic
variety over the field of complex numbers, then the space X(~) is con-
nected. By covering X with affine open sets (in the spectral topology) it
is easy to reduce the theorem to the case when X is affine. In that case
the proof is based on the fact that according to Theorem 10 in Ch. I, § 5,
there exists a finite mapping f: X --+ An onto an affine space. Using the
terminology introduced in Ch. V we shall speak in what follows of finite
morphisms. In § 2.1 we deduce some simple topological properties of
algebraic varieties, and § 2.2 is devoted to a proof of the main result: the
fact that X(~) is connected. Here we make use of some simple properties
of analytic functions of several complex variables, which are proved in
§2.3.
1. Auxiliary Lemmas
Lemma 1. If X is an irreducible algebraic variety, and Y a proper sub-
variety, then the set X«[:) - Y«[:) is everywhere dense in X«[:).
We consider first the case when X is an algebraic curve. Then Y con-
sists of finitely many closed points. Let v: Xv -+ X be the normalization
morphism, Y' = v-l(y). Since Xv is a smooth curve, every point y' E Y'
has a neighbourhood U homeomorphic to the disc \z\ < 1 in the plane
of the complex variable z. Obviously U - y' is everywhere dense in U,
and therefore also XV«[:) - Y' is everywhere dense in XV«[:). Since v is an
epimorphism, it follows that X«[:) - Y is everywhere dense in X«[:).
The general case can be reduced to the one just considered by a simple
induction on the dimension n of the variety X. Suppose that n >.1. For
every point y E Y«[:) there exists an irreducible subvariety X' of co-
dimension 1 in X that contains y and does not contain any irreducible
component of Y passing through y. For we take in an affine neighbour-
hood U of y a point Yi =1= y on every irreducible component Yi of the
variety Y passing through y, and we consider the section of U by a
hyperplane L in the enveloping affme space such that y E L, Yi ¢ L, for
all the chosen points Yi. The closure in X of any irreducible component
ofthis section passing through y can be taken for the variety X'. We set
Y' = X' n Y. By the inductive hypothesis X'«[:) - Y'«[:) is everywhere
dense in X' «[:). In particular, the point y lies in the closure of
X' «[:) - Y' (<r). Therefore, a fortiori, it lies in the closure of X «[:) - Y «[:).
Since y can be taken to be any point of Y«[:), this proves the lemma.
Corollary. If X is an irreducible algebraic variety, Y =1= X an algebraic
subvariety, and X«[:) - Y«[:) is open in X«[:) and connected, then X«[:) is
also connected.
For if X«[:) = Ml U M2 is a decomposition into two closed disjoint
sets, then X«[:)- Y«[:) splits into its intersections with Ml and M 2 •
Since X«[:) - Y«[:) is connected, it must coincide with one of these
intersections, hence be contained in Ml or in M 2 • But then also its closure
is contained in the corresponding set. According to Lemma 1 this closure
coincides with X«[:), and this means that one of the sets Ml or M2 is
empty.
Lemma 2. If V CJAn is closed in the spectral topology, then V«[:) is
connected. .
Proof. Let JAn - V = Y, Xl' X2 E V«[:). Through Xl and X 2 we draw a
line L that does not contain any irreducible component of Y. The space
L«[:) is homeomorphic to ([:, and L«[:)n V«[:) to the space([: - {Yl'·· .,Ym},
where {Yl' ... , Ym} = Ln Y. It follows that L«[:)n V«[:) is connected, hence
that Xl and X2 belong to one and the same connected component of
V«[:). Since Xl and X2 are arbitrary, V«[:) is connected.
2. The Main Theorem. The proof that the space X(<C) is connected is
based on the following result, which reduces the problem to a simpler
situation.
Lemma. For every irreducible variety X there exists a set U C X, open
in the spectral topology, and a finite morphism f : U --+ V onto an open
subset Vof the affine space IN (in the spectral topology) such that the
following conditions hold:
1) U is isomorphic to a subvariety V(F) in V x JA l , and
F(T) E <C[JAn] [T] C <C[V x JAl]
is an irreducible polynomial over <C[JAn] with highest coefficient 1, and
f : U --+ V is induced by the projection ·v x JA 1 --+ V;
2) the continuous map f : U(<C)--+ V(<C) is an unramified covering.
Proof. Let X' C X be an affine set that is open in the spectral topology,
and let f : X' --+JAn be the finite morphism whose existence is guaranteed
by Theorem 9 in Ch. I, § 5. In the ring <C[X'] ) <C[JAn] we consider a
e
primitive element of the extension <C(X')/<C(JAn), and we denote by F(T),
FE <C[JAn] [T], its minimal polynomial' over <C(JAn). Evidently,
<C[JAn]le] C<C[X'], and from the fact that e is a primitive element it
follows that for every element ~ E <C[X'] there exists an element a~ E <C[JAn]
such that oc~· ~E<C[JAn] [e]. If ~l' •.. , ~s is a basis of the module <cCX']
over <C[JAn] and a=a~I ... a~s' then
a . <C[X'] C<C[JA1 [e] . (1)
We denote by dE <C[JAn] the discriminant of the polynomial F(T),
we set c = a . d, V = JAn....: V(c), U =X' - V(c), and as before, we denote
by f : U --+ V the restriction of the morphism f. We show that for U, V,
and f the conditions 1) and 2) of the lemma are satisfied.
Obviously, by virtue of (1) <C[U] = <C[X'Jc = <C[JAn] [eJc=<C[V] [e].
Therefore U is isomorphic to the subset V(F) in V x JA l , and 1) holds.
Let m = degF, Vo E V. Since d(voh~: 0 (d is the discriminant of F), the
equation F(vo)(T) = 0 has m distinct roots J.tl' ... , J.tm' and
(2)
We denote by Zl' ... , Zn coordinates in JAn. They determine local param-
eters at Vo EJAn. By (2) they [more accurately, f*(Zl)' ... ,f*(zn)] deter-
mine local parameters also at U j = (vo, J.tJ Therefore, there exist neighbour-
hoods OJ of Uj and V of Vo in the complex topology such that the mappings
U--+(f*(Zl)' ... ,f*(zn» and V--+(Zl' ... , zn) are homeomorphisms of 0
and V onto one and the same domain of <cn. This shows that f : OJ --+ V
is a homeomorphism. We can take V so small that the OJ do not intersect
for i=l=j. Then f-l(V)=UOj. For the number of inverse images of a
point v E V cannot exceed m, but it has already m inverse images in

Ul , ... , Urn· This completes the verification of condition 2).*
Theorem. If X is an irreducible algebraic variety over the field of
complex numbers, then the space X(<C) is connected.
Proof Let U be the set whose existence is guaranteed by the lemma.
According to the Corollary to Lemma 1 we need only prove that the set
U(<C) is connected. Let U(<C) = Ml U M2 be a decomposition into two
closed disjoint subsets. The mapping f: U(<C)---+ V(<C) constructed in the
lemma carries open sets into open sets and closed sets into closed sets.
Since Ml and M2 are open and closed in U(<C), we see that f(M l ) and
f(M2 ) are open and closed in V (<C). By LemIila 2, V(<C) is connected,
hence f(M l ) = f(M2 ) = V(<C).
Clearly the restriction of f to Ml determines an unramified covering
it: Ml ---+ V(<C). From the fact that V(<C) is connected it follows easily
that the number of inverse images in Ml of a point v E V(<C) is one and
the same for all v. We denote this number, the so-called degree of the
covering f : Ml ---+ V(<C), by r. Since also f(M2 ) = V(<C), we have r < m,
where m is the degree of the covering f.
For a point v E V(<C) we take a neighbourhood Vv such that it -l(Vv)
= Ul U ... U Ur, Ui n Uj = 0 for i i= j and that the restriction of it to Ui is
a homeomorphism fi: Ui ---+ Vv, i = 1, ... , r.
For every function 8 E <C [U] that is integral over <C [An] we consider
its restrictions 8 1 , ... , 8r to the sets Ul , ... , Ur, and we denote by gl' ... , gr
their elementary symmetric functions. The idea of the next argument is
as follows: we show that there exist polynomials PI' ... , Pr E <C[.An ] such
that at all points v E f (Ml ) their restrictions to Vv coincide with g 1, ... , gr'
Hence it follows easily that 8 satisfies a relation
(3)
at all points x E MI'
Since a similar relation (with another r' < m) holds at the points
x E M2 there exist polynomials PI' P2 E <C[E] [T] of degree less than m
such that ~(8)=O in M;, i=1,2. Therefore Pl (8)P2 (8)=O in <C[U],
and since <C [U] has no divisors of zero, we see that the function 8 E <C[U]
satisfies an equation over <C [An] of degree less than m. This contradicts
the fact that by definition m ~ [<C( U) : <C(An)].
Proceeding to putting this plan into effect we observe, first of all,
that the functions (fi- l )*(8) are analytic on the set Vv in the coordinates
Zl' ... , Zn in An(<c). For according to the lemma, the local parameters at
ui=fi-l(V) can be expressed as analytic functions of f*(Zl), ... ,f*(zn)'
* The verification of condition 2) could, of course, be replaced by a reference to
Ch. II, § 5.3. We have preferred to give an independent simple proof, to avoid using the
more difficult Theorem 8 in Ch. II, § 5.
and () is in a sufficiently small neighbourhood an analytic function of the

local parameters at v. Thus, g1' ... ,gr are also analytic functions of
Z1' ... , Zn in VV ' Consequently, each of the functions gi is analytic in
Z1' ... , Zn on the whole set V(<c). We recall that V(<C) is obtained from the
whole space N(<C) by excluding points of an algebraic subvariety S C /An.
We consider the behaviour of the functions gi in a neighbourhood of a
point s E S(<C). By the choice of () it satisfies an equation
The values of the functions U;- 1)* ((}) are roots of this equation. Therefore,
in every compact neighbourhood of s the functions gi are bounded. From
this it follows that they may be extended to functions analytic in the
whole of /An (<C) (Lemma 1 in § 2.3).
We show that the so obtained analytic functions gi on /An(<c) are
polynomials in the coordinates Z1' ... , Zw
For this purpose we estimate the order of their growth in dependence
of the growth of max\zJ For a point Z=(Z1, ""Zn)E/An(<c) we set
\z\ = max \z;\. Applying to the equation (2) the standard estimate for the
modulus of a root of an algebraic equation we find that
\(}(x)\ < 1 + m~x \ai(f(x))\.

!
By hypothesis, the functions ai are polynomials in Z1' ... , Zw If the

maximum of the degrees of these polynomials is equal to I, then there
exists a constant C such that
\(}(x)\ < ClZ\1 .
Hence it follows that (fi- 1 )*(8) for every i= 1, ... ,r satisfies the same
inequality and therefore that
\gi(Z)\ < ClZ\il, i = 1, ... , r.
So we see that the gi(Z) are analytic functions in the whole of /An (<C) = <cn
having polynomial growth. Consequently they are polynomials in
Z1' ... , zn (Lemma 2 in § 2.3).
So we have established the relation (3), and hence have concluded the
proof of the theorem.
3. Analytic Lemmas. We prove here the two lemmas on analytic

functions of complex variables that were used in the preceding subsection.
Lemma 1. Let S C /An, S #- /An be an algebraic subvariety and g an
analytic function on the set /An(<c) - S(<C) that is bounded in the neighbour-
hood of every point s E S(<C). Under these conditions g can be extended to a
holomorphic function on the whole of .i\n(<c) = <cn , and there is only one
such extension. .
The uniqueness of the extension follows at once from the uniqueness
theorem for analytic functions.
Clearly it is sufficient for us to find for every point s E S(<C) a neigh-
bourhood U such that g can be extended from U - (U nS(<C)) to U as an
analytic function. Then we obtain a single extension by virtue of its
uniqueness.
To prove the existence of the extension we observe that we may
replace S by a larger algebraic subvariety and may therefore assume
that S is given by a single equation f(Zl, ... ,zn)=O. Making the trans-
formation zi = Zi + CiZm i = 1, ... , n - 1, z~ = Zm for suitably chosen
numbers c 1, ... , Cn- 1 we can achieve that f as a polynomial in Zn has
highest coefficient 1.:
f(Zl, ... , zn) =z~ + h 1(z') Z~-l + '" + hz(z'),
where Z'=(Zl, ""Zn-1)'
Suppose that s is the origin of coordinates. Then
f(O, ... ,0, zn) = z::'(zn - AI), ... , (zn - An-m) .
By the theorem on the continuity of the roots of an algebraic equation
the roots of f(z';zn)=O tend, as z'~O, either to 0 or to AI' ... ,An- m.
Therefore, we can find a real number r> 0 and an e > 0 such that for
Iz'l < e the equation f(z'; zn) = 0 does not have roots with IZnl = r. We
define Iz'l = max IzJ
1, ... ,n-1
We set
1.
G(Zl, ... ,Zn)=-2 J
g(z';w) dw
7rl Iwl~r W-Z n
and show that G is an analytic function for Iz'l < e, IZnl < r, and is an
extension of g to this domain.
The fact that G is analytic can be verified directly by integration. By
assumption, g is analytic at every point (tx 1 , ... , txn-1, w), Itxil < e, Iwl = r.
~
Th erelore th e functIOn
. g(zl"",Zn-1;W) for every w WIt 'h Iw I = r IS .
w-z n
analytic at the point Zl =tx 1 , ""Zn-1 =txn- 1 , zn=[3, 1[3I<r. Integrating
its expansion in a Taylor series
g(Zl"",zn-l;W)
W -Zn
= Le ll , ... , zjw)(z 1 - tx 1 )z\ ... , (zn-1 - tx n_ 1)Zn -I(zn - f3)ln (1)
with respect to the circle Iwl = r, we obtain the Taylor expansion for G.
We now show that G agrees with g where g is defined. For this purpose
we set Zi = ai' i = 1, ... , n - 1, lail < 8, and we consider the functions
g(a; zn) and G(a; zn). The preceding arguments show that G(a; zn) is
analytic for IZnl < rl < r, and g(a; zn) is by hypothesis analytic at all
points Zn with IZnl < rl, except possibly the finitely many roots of the
equation f(a; zn) = 0, but is bounded in their neighbourhood. Therefore,
it has no poles for IZnl < r, and formula (1) shows that G(a; zn) = g(a; zn)
by Cauchy's integral formula.
Lemma 2. Let f(Zl' ... , zn) be an analytic function in the whole of(Cn
and suppose that there exists a constant C such that
(2)
Then f is a polynomial of degree .;:;; 1.

Suppose that in the expansion of f in a Taylor series
a homogeneous constituent Fj for some j> I is not identically O. We

can find numbers a 1 , ••• ,an such that Fj(a 1 , .•• ,an);60. Then for g(w)
= f(a 1 w, ... , an w) the coefficient of wj in the Taylor series is non-zero,
and, as before, the estimate (2) holds. Subtracting from g the sum of the
first I terms of the Taylor series we obtain the function
for which az ;6 0 and (2) is also true.

By hypothesis, the function g dwz is bounded in the whole plane,
hence constant. This contradicts the fact that az ;6 O.
Exercises
1. Show that if X is a quasiprojective variety and the space X(<C) is compact, then X
is projective.
2. Show that if the space X(<C) is compact, then the variety X is complete. Hint: Use
Exercise 7 to Ch. VI, § 2.
3. Let X be a reduced and irreducible scheme of finite type over <C, and X(<C) the set
of its closed points equipped with the same topology as in § 1.1. Show that if X(<C) is a
Hausdorff space, then X is separated.
4. Show that the automorphism group of anon-hyperelliptic smooth projective curve
of genus oF 0 is finite. Hint: Show that if cp : X --> lPg- 1 is the embedding corresponding to
the canonical class, then the automorphisms of the curve cp(X) are induced by projective
transformations in pg-l and therefore form an algebraic group G. Apply Exercise 4 to § 1.
5. Extend the result of Exercise 4 to hyperelliptic curves.
§ 3. The Topology of Algebraic Curves

The association of the topological space X(<C) with an algebraic variety
X leads to problems of two types. Firstly, it is interesting to clarify what
topological spaces we obtain in this manner, and if possible, to go as
far as a topological classification of them. Secondly, to investigate which
of the invariants of the topological space X(<C) have an algebraic meaning.
In other words, it is a question of constructing for a variety X defined
over an arbitrary field k invariants which for k = <C would tum into
prescribed invariants of X(<C).
In this section we indicate how to solve problems of the two types in
the simplest case, when X is a smooth projective curve. However, this is
almost the only case in which a topological classification of the spaces
X(<C) and the algebraic meaning of the resulting topological invariants
are completely known.
1. The Local Structure of Morphisms. Let X be a smooth algebraic

curve. Every point x E X(<C) has a neighbourhood U (in the complex
topology) that is homeomorphic to a neighbourhood of the origin of
coordinates in the complex plane <C. This homeomorphism can be given
by any local parameter t at x:
t: U-<C. (1)
From now on we assume U to be chosen so that t(U) is the domain

Izl < e in <C. The mapping (1) allows us to regard t as a coordinate in U
so that the number t(x), x E U, determines the point x uniquely.
Let f : X - Y be a morphism of smooth curves, f (X) dense in Y, x E X,
f(x) = y. We show that the points x and y have neighbourhoods U and V
and in them coordinates u and v in which the mapping f has a very
simple expression.
We choose local parameters t and v at the points x and y and neigh-
bourhoods U of x and V of f(x) such that f(U) c V and the open embed-
dings (1) hold:
t: U-<C,
v: V -<C.
Regarding t as coordinate on U and v on V, we can say that f is defined
in U by prescribing v(j(x'») as a function of t(x ' ) for points x' E U. In
other words, to specify f we have to represent f*(v) as a function of t.
Since f(X) is dense in Y, the function f*(v) does not vanish identically
on X, is regular, and vanishes at x. We set
(2)
To q> there corresponds a formal power series 4>(T), 4>(0) =F- 0, with a
positive radius of convergence. Therefore, by taking, if necessary, for U a
smaller neighbourhood of x we may assume that
q>(x') = 4> (t(x')) , x' E U.
Since 4>(0) =F- 0, there exists a power series 1p(T) = 4>(T)llf, also having a
positive radius of convergence. Therefore the function u(x') = 1p(t(x'))· t(x')
is defined for points x' in a sufficiently small neighbourhood of x, which
we denote again by U. So we have constructed a mapping
u:U-+<C,
with
The function u = t· P(t) is no longer rational on X and is defined only

in a sufficiently small complex neighbourhood U of x. However, in this
neighbourhood it is evidently continuous. Like t, it determines a homeo-
morphism of some neighbourhood of x onto an open set in <C. For by
the implicit function theorem the analytic function z P(z) has in some
neighbourhood of 0 an inverse function and therefore determines a
homeomorphism of some neighbourhood of 0 in <C.
By construction, f*(v) = U Z in the open set U. Thus, we have reached
the end of our analysis by obtaining a simple local representation for the
mappingf.
Theorem 1. For every morphism of smooth curves f : X -+ Y for which
f(X) is dense in Y, and for every point x E X there exist neighbourhoods U of
x and V of f(x) and homeomorphisms u: U -+<C and v: V -+<C onto neigh-
bourhoods of 0 in <C such that
is commutative. Here Qz(z) = Zl, and 1 is defined as the order of the zero
of f*(t) at x if t is a local parameter at y.
If u and v are interpreted as coordinates in U and V, then Theorem 1
asserts that under restriction to these sets and in these coordinates the
mapping f can be expressed in the very simple form
(3)
Obviously, the open sets V and V can be chosen so that u(U) and v(V)
coincide with the open disc Izl < 1.
Such neighbourhoods are called normal.
The number I in (2) and (3) is called the order of ramification of fat
the point x E x. If the order of ramification of f in at least one point
xEf-l(y) is greater than 1, then y is called a ramification point of f.
Clearly, the order of ramification of f at x is the same as the
multiplicity with which x occurs in the divisor f*(y). If the curves X
and Yare not only smooth, but also projective, then Theorem 1 of
Ch. III, § 2, shows that y is not a ramification point if and only if the
number of inverse images of this point is equal to the degree degf of the
morphism f. In other words, our definition is compatible with the
definition of a ramification point given in Ch. II, § 5.3. From Theorem 7
in Ch. II, § 5, it follows that a morphism of smooth projective curves has
finitely many ramification points.
2. Triangulation of Curves. Here we show that the space X(CC) is

triangularizable, where X is a smooth projective algebraic curve. For the
convenience of the reader the definition of triangulation and the basic
facts on the classification of triangularizable two-dimensional manifolds
are compiled in § 3.4.
The triangularizability of the space X(CC) is obtained as a consequence
of a more general fact. To state it we introduce the following definition.
A triangulation ,p of a topological space X is said to be compatible
with a triangulation tp of a space Y relative to a continuous mapping
f: X -+ Y if f -l(E) = UEi' Ei E,p, for every simplex E E tp, and if the
mapping f : Ei -+ E is a homeomorphism.
Theorem 2. If f : X -+ Y is a morphism of smooth projective curves and

if the space Y (CC) is triangularizable, then X (CC) and Y (CC) have triangulations
that are compatible relative to f. If Y(CC) is a combinatorial surface, then
so is X(CC).
Proof. We consider an arbitrary point yE Y(CC), and we set f-l(y)
= {Xl' ... , Xl}. According to Theorem 1 we can take a neighbourhood Vy
of y and neighbourhoods Vi of the points Xi that are normal and pairwise
disjoint.
From the covering Vy of Y(CC) we select a finite subcovering. We
obtain a finite set of points Yex in Y(CC), for each point Yex a neighbourhood
Vex, and for Xex,i E f -l(yex) normal disjoint neighbourhoods Vex,i. Obviously,
if y E Vex, y:;6 Yex' then y is not a ramification point.
By assumption Y(CC) has a triangulation tp o. Bya proposition to be
proved in § 3.4 there exists a finer triangulation tp such that all ramifica-
tion points of the morphism f are vertices of it and that every simplex
of it is contained in some neighbourhood Va'
Let E be an arbitrary simplex of the triangulation P. By hypothesis,
E is contained in some open set Va' If Ya is not a ramification point, then
J;: Ui,a~ Va for every XiEf-I(yJ is a homeomorphism. We denote by
Ei the inverse image f - 1 (E) in Ui,a' If t: E ~ (J' is the mapping that occurs
in the definition of the triangulation P, then we set ti = tfi: Ei~(J'. We
include the sets Ei and the homeomorphisms ti in the triangulation iP.
Suppose now that E C Va and that Ya is a ramification point. We
consider two cases.
a) Ya ¢ E. In suitable coordinates the mapping fi: Ui,a ~ Va has the
form v = f *(u) = U /. Since Ya ¢ E and the set E is simply-connected, we
see that v 1= 0 on E and that every branch of the function Vv determines
there a single-valued function. From this it follows that fi - 1 (E) splits
into I connected components E 1 , ... , El and that the mapping fi: Ei ~ E
is a homeomorphism. We include the sets Ei and the mappings
tj= tfj: Ej~(J' in the triangulation iP.
b) Ya E E. Now we can apply the same arguments to the set E - Ya'
We find that f- 1 (E-Ya) splits into I connected components £1,,,,,£/,
Setting
we see that fj: Ej~ E is a homeomorphism. Again we include all the E j

and the mappings tj=tfj:Ej~(J' in the triangulation iP.
A simple verification shows that the sets and mappings so constructed
determine a triangulation iP of the space X(<C), and also that this triangula-
tion is compatible with P relative to f and satisfies the definition of a
combinatorial surface. This verification is left to the reader.
Theorem 3. If X is a smooth projective curve, then the space X(<C) is
triangularizable and is a combinatorial surface.
We show first the triangularizability of the space ]pI (<C). We indicate
a special triangulation, which is not very economical, but is useful for
subsequent applications.
The decomposition of the surface of an octahedron into faces, edges,
and vertices gives a triangulation of it. We imagine the octahedron to be
inscribed in a two-dimensional sphere. By projecting this triangulation
from an interior point of the octahedron we obtain a triangulation of the
two-dimensional sphere. Since the projective line over the field of complex
numbers is homeomorphic to a two-dimensional sphere, we obtain a
triangulation of the projective line.
We identify]P 1 (<C) with the plane of a complex variable, supplemented
by the point at infinity. The triangulation we have constructed is given by
the decomposition corresponding to the real axis, the imaginary axis,

and the circle Izl = 1.
Fjo
~l --::G:-I3f-----"----l-:::G-
s ----=---+.,,-G1- Fg
d(E;) = 2, i = 1, ... ,8; d(F;) = 1 ,

i=1, ... ,12;d(G;)=O, i=1, ... ,6.
Now we consider a non-constant rational function f on X. It deter-

mines a morphism f : X ~ ]p 1. It remains to apply Theorem 2.
3. Topological Classification of Curves. We apply the topological

classification of surfaces that will be explained in § 3.4 to the surfaces
X(<C), where X are smooth projective curves. According to Theorem A of
§ 3.4, for this purpose we have to clarify whether the spaces X(<C) are
orientable, and we have to compute their Euler characteristics.
Theorem 4. If X is a smooth projective curve, then the space X(<C) is

orientable.
This is, of course, a special case of Proposition 1 in § 1.2; however,
we give another much more elementary proof.
We use an arbitrary morphism f : X ~ ]p 1 and consider triangulations
iP and '1' of the spaces X(<C) and ]Pl(<C) that are compatible relative to f.
Their existence follows from Theorem 2. Since the two-dimensional
sphere is orient able, so is the triangulation '1'.
From the fact that iP and '1' are compatible it follows that the simplexes
of iP are precisely those into which the inverse images of simplexes in '1'
decompose. Therefore every simplex E E iP is mapped by f homeo-
morphically onto a simplex FE '1'.
We fix an orientation of 'P, that is, of all its triangles. To a triangle

E E <P we give the orientation that is obtained from the orientation of
f(E) E 'P by means of the homeomorphism f: E--+ f(E). It remains to
verify that in this way we obtain an orientation of the whole triangula-
tion <P. This is simple. Suppose that two triangles E' and E" have a
common edge E with the vertices band c. We denote the vertices of E'
and E" by a, b, c and b, c, d. We set f(a) = ai, f(b) = b', f(c) = c', f(d) = d' .
Then ai, b', c' and b', c', d' are the vertices of the triangles f(E') and f(E")
of 'P. Suppose that the chosen orientation gives for f(E') the order
(ai, b', c' ) of the vertices. Then according to the definition ofthe orientation
ofa triangulation, the order in f(E") must be (c', b', d' ). By hypothesis, the
order of the vertices in E' and E" is (a, b, c) and (c, b, d), hence it is clear
that.on E they determine opposite orientations. This proves the theorem.
Now we proceed with the second problem: the definition of the
Euler characteristic of X (<C) for an irreducible smooth projective curve X.
Theorem 5. The Euler characteristic of the space X(<C) is 2 - 2g,
where g is the genus of X.
Again we use a regular mapping f: X --+]p I and compatible tri-
angulations <P and 'P of the spaces X(<C) and ]P1(<C). We denote the
numbers in the Definition (2) of § 3.4 of the Euler characteristic by
co, c l , C2 for <P and c~, C'l' C2 for 'P. Then
Let us see how these numbers are connected. By the definition of com-
patibility of triangulations, for every simplex E E 'P
(1)
where f : E;--+ E is a homeomorphism. In the proof of Theorem 4 we have

seen that by sorting out all the simplexes E E 'P we obtain among the E;
all the simplexes of <P.
How many simplexes E; are there in (i)? If degf = nand d(E) > 0,
then there are n of them. For their number cannot be larger than n,
because every point Y E]P1 has at most n inverse images. But it cannot
be smaller than n, because there are only finitely many points (the rami-
fication points) having fewer than n inverse images. Let d(E) = 0, so that
E is a point Y E ]p I (<C). Denoting by y the divisor consisting of the single
point y with multiplicity 1 we set
r
f*(Y) = L l;x;. (2)
;= I
r
The number of inverse images of y is r, but since L li = n, we have
i= 1
r = n - L (li - 1). Therefore we obtain
C2=nc~,
c 1 =nc'l,
Co =nc~ - L(l i -1),
where the last sum contains the ramification orders of f at all points. As
a result we see that
On the other hand, from the triangulation (1) of § 3.2, for example, it is
clear that X(]Pl(<C)) = 2. Therefore finally
X(X(<C)) = 2n - L(li - 1) . (3)
We now consider an arbitrary rational differential form w =1= 0 on ]pl

and compute the divisor (f*(w)) ofthe inverse image of won X. Suppose
that at YE]pl we have vy(w)=m. Then
w= tm. g' dt,
where t is a local parameter at y, g E (!Jy, g (y) =1= O. If the numbers Ii denote

the same thing as in (3), then vx.(f*(t)) = Ii' that is, f*(t) = '[Ii h;, where '[ is
a local parameter at Xi' hi E (!JXi' h;(Xi) =1= O. From this it follows that
In other words,
(f*(w)) = f*((w)) + L(li - 1) Xi' (4)
where, as in (3), the last sum is extended over all points Xi E X at which
the ramification exponent of f is greater than 1.
Since f*(w) is a differential form on X, we have deg((f*(w)))=2g-2.
In exactly the same way deg((w)) = - 2. Finally, for every divisor D on
]pl the equation deg(f*(D)) = n degD holds, because this is true for a
divisor consisting of a single point, by Theorem 1 of Ch. III, § 2. Con-
sidering the degrees of the divisors on the two sides of (4) we therefore
find
Comparing this formula with (3) we obtain Theorem 5.

Theorems 4 and 5 in conjunction with the topological theorem A of
§ 3.4 give us a complete topological classification of smooth projective
curves. They show that for two such curves the spaces X(<C) are homeo-
morphic if and only if the curves have the same genus.
No similar result is known for varieties X of dimension > 1. We
quote one of the simplest results on the connection between topological
and algebraic properties of smooth complete varieties, which generalizes
Theorem 5. Since X(X(<C)) = bo - b i + bb where bi is the one-dimensional
Betti number of X(<C) and b o = b 2 = 1, Theorem 5 can be expressed by the
equation
If X is an arbitrary projective smooth variety, then a similar result

holds:
where b i is the one-dimensional Betti number of the space X(<C) and

hI = dim<e.Q I [X]. Using more delicate constructions we can express all
the remaining Betti numbers of X(<C) in terms of algebraic invariants
of the variety X.
In conclusion we mention that the topological classification of
smooth projective curves can be obtained by other means, within the
framework of the theory of differentiable manifolds. Then we have to
use a less elementary topological apparatus, but on the other hand, the
exposition is more invariant. We sketch only the general course of this
exposition, omitting all details.
In § 1.2 we have seen how to prove orientability on the space X(<C) for
every smooth variety X, by using concepts of the theory of differentiable
manifolds. The combinatorial classification of surfaces (Theorem A of
§ 3.4) must be replaced by its "smooth" analogue, the theorem that every
connected compact orient able surface can be obtained by attaching
finitely many handles to the sphere. The proof of this theorem follows
easily from Morse theory (see, for example [35]). It remains to prove
Theorem 5. For this purpose we have to consider the tangent bundle 8
of the two-dimensional surface X (<C). Its first Chern class C 1 (8) is an
element of the group H 2 (X(<C), Z). This group has a canonical generator:
the homology class cp for which cp(w x ) = 1, where Wx is the orienting
class of X. Therefore C I (8) = v· cp, V E Z, hence the integer v is determined.
In our class C I is the Euler class, and therefore v = X(X(<C)), that is,
c I (8) = X(X(<C))· cp, where X(X(<C)) is the Euler characteristic of X(<C)
(see [19], 388).
On the other hand, as an algebraic vector bundle over X, (J is one-

dimensional and corresponds to the divisor class - K, where K is the
canonical class of X. Since deg( - K) = 2 - 2g, the relation
X(X(<C») = 2 - 2g
follows from the general result.
If E is a one-dimensional vector bundle on a smooth projective curve
X, D its characteristic class, and c I (E) the Chern class of the corresponding
bundle on X(<C), then
cI(E)= (degD)/p .
A proof of this for the case of a divisor D on a variety X of arbitrary
dimension is given in [11].
4. Combinatorial Classification of Surfaces. For the convenience of

the reader we recall here some elementary topological concepts and
results.
Let V be an n-dimensional affme space over the field of real numbers.
Then any two points P, Q E V determine the vector PQ belonging to the
n-dimensional vector space IRn, and ~ery vector x E IRn and point P E V
determine a point Q E V such that PQ = x; this is written as P + x = Q.
For any points PI' ... ,PmE V and numbers AI' ... ,~EIR such that
!;Ai = 1, the point Q + !;Ai~ does not depend on the choice of the
auxiliary point Q and is written in the form !;AiPi. If the points Po, ... ,
Pr E V do not lie in any affine subspace of dimension less than r, then the
r
representation of R in the form R = L A;1~, !;Ai = 1, is unique.
i=O
A set of points REV that are representable in the form
r
R= L AiPi> L Ai=1, Ai~O, (1)

i=O
in terms of independent points Po, ... , Pr is called an r-dimensional

simplex. The points Pi are called its vertices.
If p;" ... , Pir - s are any r- s vertices of a simplex a, then the points
R E a for which Ai, = ... = Air _ s =Oin(1)themselves form an s-dimensional
simplex with the vertices Pi' j =1= iI' ... , i r - s • This simplex is called a
face of a.
Let X be a Hausdorff space. We define a triangulation (more
accurately, a finite triangulation) of X. This is the name for: a) a finite
family 
where ai is a simplex of dimension d(Ei). Here the following conditions
must be satisfied:
1) X=UE i ;
2) if Ei E <P and Ej E <P, then EinEj E <P or is empty;
3) if E j C E i, then ti(Ej) is a face of (Ji; all the faces of this simplex are
obtained in this way.
From the definition it follows that if d(Ei)=O, then Ei is a point
x E X. All these points are called the vertices of the triangulation. The
subsets Ei are called a simplicial triangulation, and the vertices of the
triangulation contained in a given simplex Fi are called its vertices.
It is easy to show that if the set K={x 1 , ... ,XN} of the vertices ofa
triangulation is known to us and also what subsets S C K are the sets of
vertices of a single simplex, then we can reconstruct X from this informa-
tion. Thus, a triangulation of a space enables us to give a purely com-
binatorial scheme for it. Topological spaces admitting at least one
triangulation are called triangularizable.
In connection with triangulations of the spaces X(<C), where X is a
smooth projective curve, we need triangulations <P having the following
properties:
a) all the simplexes of the triangulation are of dimension ,;;; 2,
b) every simplex of dimension < 2 is a face of some simplex of
dimension 2,
c) every simplex of dimension 1 is a face of precisely two simplexes
of dimension 2.
A topological space having a triangulation with these properties is
called a combinatorial surface.
In what follows we make use of the operation of refinement or sub-
division of a given triangulation. We give a simplified description of the
operation of subdivision that fits triangulations for which d(E i ),;;; 2 for
all Ei E <P.
Let X be a topological space, <P a triangulation of it with d(Ei)';;; 2
for all Ei E <P, and Er one of the simplexes of the triangulation for which
d(E r} = 1. We choose any interior point ~ on the segment tr(Er) and
denote by r' and r" the parts into which ~ divides this segments. We
set x = tr- 1 (~). Let E i, i E I, be those simplexes of <P for which d(Ei) = 2,
E i ) Er. We divide the triangle ti(E i) into two triangles T;' and T;" by
joining the point ti(x) to the vertex opposite the side ti(E,).
We consider the family <P' consisting of the following closed subsets
of X:
d=O: those EjE <P for which d(Ej)=O, and the point x:
d = 1: those Ej E <P for which d(E) = 1, j =1= r, and t; 1 (r'), t; 1 (r");

d = 2: Ej E <P, j¢ I, and t i- 1 (T'), ti l(T") for i E I.
For example:
Thus, every set of the family rp' either coincides with a simplex of the
triangulation rp or is part of such a simplex. We define a mapping t' as
the corresponding mapping in rp or its restriction.
It is easy to verify that the conditions 1), 2), and 3) hold, so that rp'
is a triangulation. It is called a subdivision of rp.
The following property is an immediate consequence of the definition
of subdivision.
Proposition. Let SeX be a finite set, X = UUi a finite covering, and rp
a triangulation of X. There exists another triangulation obtained from rp
by a sequence of subdivisions for which all the points of S are vertices and
each of its simplexes is contained in one of the sets Ui.
Such a triangulation is said to be inscribed in the covering {U;}.
If a point s E S belongs to a simplex Ei with d(El) = 1, then by a
single process of subdivision we make it into a vertex. But if sEE;,
d(EJ = 2, then we choose a simplex E j C E;, d(E) = 1, and the vertex P
of the triangle ti(E i) opposite the segment ti(EJ For ~ we choose the
point of intersection of the side ti(E) and the line joining ti(s) and P. By
performing the corresponding subdivision we arrive at the previous case.
In order to inscribe a triangulation in a covering it is sufficient to
do this for every triangle. We leave the details of this simple argument to
the reader.
Now we recall the concept of orientation of some triangulation of a
surface. An orientation of a triangle or a segment is a choice of one of the
two possible directions of going around its vertices. Thus, every triangle
or segment has two distinct orientations, which are called opposite.
Every orientation of a triangle determines an orientation of is edges.
Let X be a combinatorial surface and rp a corresponding triangulation.
An orientation of rp is a choice of orientations of all its triangles such that
on every segment E E rp, d(E) = 1, the triangles whose face E is determine
opposite orientations. A triangulation having an orientation is said to be
orientable. If X is connected, then a triangulation rp has either precisely
two orientations or none. It is easy to verify that the triangulation of the

two-dimensional sphere [or of1P I (<C)J constructed in § 3.3 is orientable.
The property of the surface of being orientable does not depend on
its triangulation. In other words, if <P and P are two distinct triangulations
of one and the same surface, then they are either both orientable or both
non-orientable. In invariant terms the condition of orientability of a
surface can be written as Hz (X, Z) #- o.
The last topological concept we need is that of the Euler characteristic
of a surface. If a triangulation <P has Co vertices, c 1 edges, and Cz triangles,
then its Euler characteristic is defined as:
Xq,(X) = Co - C1 + Cz • (2)
Like the property of orientability, the Euler characteristic does not

depend on the triangulation of the surface and is therefore denoted by
X(X). An invariant definition of it is:
x(X) = dimK Ho(X, K) - dimK HI (X, K) + dimK Hz (X, K),
where K is an arbitrary field of characteristic o. It is easy to check that

the Euler characteristic of the two-dimensional sphere [that is, of1P 1 (<C)J
is equal to 2.
The main result of the topology of surfaces is that the topological
invariants we have introduced: the property of orientability and the
Euler characteristic, form a complete system of topological invariants on
connected triangularizable surfaces.
Theorem A. Two connected triangularizable surfaces are homeomorphic
if and only if they are simultaneously orientable or non-orientable and
their Euler characteristics are the same.
F or a proof see [2J or [28J.
The condition of triangularizibility here is superfluous: it can be
shown that every surface is triangularizible, but we do not need this.
§ 4. Real Algebraic Curves

A real algebraic curve is a scheme X over the field of real numbers IR
such that X ®JR <C is an algebraic curve. Henceforth we assume that
X ®JR<C is a smooth irreducible projective curve. As before, we denote by
X(IR) the set of closed points x E X for which k(x) = IR. In simple terms,
X is a projective smooth irreducible curve defined by an equation with
real coefficients, and X(IR) is the set of points on it with real coordinates.
As we have seen, X(JR) is a compact one-dimensional manifold.
However, it is not necessarily connected, so that the analogue of
Theorem 3 in § 2 is not true here. An example of a disconnected manifold

X(JR) has occurred in Ch. II, § 3.1.
A connected one-dimensional compact manifold is homeomorphic
to a circle. This is not hard to show directly, and for the connected
components of the manifold X(JR) it follows at once from their tri-
angularizibility, which we shall prove soon. Thus, X(IR) is homeomorphic
to a certain number of disjoint circles, so that the only topological
invariant of this space is the number of its connected components.
In this section we prove the main result, which links this topological
invariant of X(IR) with algebraic properties of the curve X:
Harnack's Theorem. If X is a smooth projective curve of genus g

defined over the field of real numbers, then the number of connected com-
ponents of the space X(IR) does not exceed g + 1.
There are several proofs of this theorem. One of them runs entirely
in the real domain. It can be found in the book [20]. We give an outline
of another proof, which is interesting in that a certain property of the
space X(IR) is deduced from its embedding in the space X(<C).
1. Involutions. In the proof of Harnack's theorem which we shall set

forth the main role is played by the mapping r that associates with every
point x E JPn(<c) the point r(x) with the complex conjugate coordinates.
Evidently r determines a homeomorphism of the topological space
JPn(<c) (however, it is not an automorphism of the algebraic variety JPn!).
Since the curve X is given by equations with real coefficients, r(X(<C))
= X(<C), and r determines a homeomorphism of the space X(<C). In
other words, r is an automorphism of the scheme X @ IR <C induced by the
automorphism of complex conjugation on <C.
As before, we shall use a triangulation of X(<C), however, now it is
convenient to choose it so that it is invariant under r, that is, together
with a simplex E it also contains the simplex r(E). Let us show that such a
triangulation exists.
Here we have to repeat the whole process of constructing a triangula-
tion of X(<C). We begin with a triangulation of JP 1 (<C) that is invariant
under r. Such a triangulation was indicated in § 3.2. It is now quite easy
to verify that the proposition on refinements of a triangulation proved
in § 3.4 can be sharpened in the sense that if c[J is a triangulation that is
invariant under r, then the refinement '1' we have constructed is also
invariant under r. For this purpose we need only arrange that by breaking
up a simplex E into two subsets E' and E" we break at the same time r(E)
into r(E') and r(E").
Finally, we choose a non-constant function f E IR(X), that is, a rational
function of the coordinates with real coefficients. The corresponding
mapping f :X(<C)~]PI(<C) obviously has the property

f(r(x» = 'I:(j(x».
It is easy to verify that the process of constructing a triangulation tp of
X(<C) that is compatible with the triangulation ~ of ]pI (<C), which was
described in the proof of Theorem 4 of § 3, leads to a 'I:-invariant tri-
angulation ~ if tp was 'I:-invariant. Thus, we have the following result.
Proposition 1. The space X(<C) has a triangulation ~ that is invariant
under the homeomorphism '1:.
Clearly the set X(1R) then consists of simplexes of this triangulation.
Proposition 2. Let E be a one-dimensional simplex of a triangulation
~ of the surface X(<C) contained in the set X(1R), and let E' and E" be the
two two-dimensional simplexes whose boundary it is. Then'l:(E')=E".
Since E' and E" are the only simplexes of ~ having the boundary E,
since ~ is invariant under 'I: and 'I:(E)=E, we have either 'I:(E')=E',
'I:(E")=E", or 'I:(E')=E", 'I:(E")=E'.
Let x be an interior point of E. At x we take a local parameter tEIR(X),
for example, the equation of the hyperplane with real coefficients passing
through x and transversal to X. Suppose that U contains x and that
t: U~<C
is a homeomorphism of U onto the disc Izl < 1 in C. We take U sufficiently

small so that of all the simplexes of ~ it intersects only E, E', and E".
Since t E lR(X), we have
t{'I:(x» ='I:{t(x» , (1)
and therefore t(En U) coincides with the real diameter of the disc Izl < 1.
So we see that U -(UnE) splits into two connected components:
U n(E' - E) and U n(E" - E). Similarly t(U) - t(E) splits into two
components: the upper and the lower semicircle. Clearly distinct com-
ponents of U - (U n E) are mapped onto distinct components of the
image. But the two semicircles are complex conjugates, hence it follows
by (1) that
'I:{U n(E' - E» = U n(E" - E).
Therefore 'I:(E')nE" is not empty, hence 'I:(E')=E".
2. Proof of Harnack's Theorem. We use the homology groups with

coefficients in 7l/271. For an arbitrary surface F the group HI (F, 7l/271)
is denoted by HI (F).
Let T I , ••• , 1/ be the connected components of the space X(1R). In the
triangulation ~ they all consist of one-dimensional and zero-dimensional
simplexes. Obviously, in the group H1 (X (<C)) they are cycles, which we

denote by the same letters.
Proposition. In the group H 1 (X(<C)) the cycles T 1 , .•• , 1/ are either
independent or are connected by the single relation
T1 + ... + 1/=0.
If the proposition were not true, then in H1 (X (<C)) there would exist
a relation
In other words,
T1 + ... + T,. = oS,
where S is a two-dimensional chain of the triangulation <P (if necessary,
we change the numbering of the cycles 7;).
The two-dimensional simplexes that do not occur in S form a chain S,
and since O(S + S) = 0, we have
(1 )
Thus, everyone-dimensional simplex occurring in a cycle 7; (i .;;; r) is a
face of one two-dimensional simplex occurring with the coefficient 1 in S
and one occurring in S.
Observe that r S is a chain of <P and that
(2)
by virtue of (1) and the fact that r 7; = 7;. Since H1 (X (<C)) is a module
over 7L/27L, (2) shows that
o(S + rS) =0 (3)
Since H2 (X(<C)) = 7L/27L, it follows from (3) that either

S+rS=S+S,
and hence
(4)
or else
S+rS=O,
(5)
S=rS.
We consider an arbitrary one-dimensional simplex E1 of <P contained
in one ofthe 7;, i = 1, ... , r. Let E' and E" be the two-dimensional simplexes
whose boundary it is. We may then suppose that E' occurs in S, and E"
in S, with the coefficient 1. By applying Proposition 2 of § 4.1 we see
that (5) is not true, so that (4) is true.
We now consider the set T,.+ 1 (by assumption, r < I) and take in it
an arbitrary point t. Clearly t E S or t E S, but t ¢ S n S, because this
intersection is Tl + ... + 1',. and does not intersect 1',.+ 1. If, say, t E S,
then -r(t) E -r(S). But -r(t) = t, because t E 1',.+ 1 C X(IR), and -r(S) = S, there-
fore t E S n S, which is not true, as we have seen. This proves the pro-
position.
To complete the proof of Harnack's theorem we need one further
topological argument. If F is an arbitrary orientable surface, then the
group Hl (F) is a module of finite rank mover Z/2Z. The intersection
index associates with two elements oc, fJ E Hl (F) an element of Z/2Z,
denoted by (oc, fJ). The function (oc, fJ) is linear in each of its arguments
and skew-symmetric, that is, (oc, oc) = 0 for every oc E Hl (F). From
Poincare's duality law it follows that it is non-singular, that is, if ~ 1,·· .'~m
is a basis of Hl (F), then
Hence it follows that arbitrary elements OCl' ••• ' OCn E Hl (F) for which
n > ml2 and (OCi' OCj) = 0, i, j = 1, ... , n, are linearly dependent.
We apply this remark to the group Hl (X (<C)). As was proved in § 3, its
rank is 2g. The cycles T 1 , ••• , T z, the connected components of the set
X(IR), are by definition disjoint. Therefore (T;, Tj)=O, i,j= 1, ... , I, hence
any g + 1 of them are linearly dependent in Hl (X (<C)). If the number 1 of
components were greater than g + 1, we would arrive at a contradiction
to the proposition. This completes the proof of Harnack's theorem.
3. Ovals of Real Curves. In his lecture on "Problems of Mathe-

matics", Hilbert, in the context of Harnack's theorem, raised the question
of the mutual disposition of the connected components of a plane real
curve X C]p2 (these components are called ovals of X). We indicate here a
precise formulation of this problem, but only for the case of curves of
even degree.
In that case it can be shown that any oval of X is homologous to 0 in
the topology of the space lP 2 (IR) and splits it into two components, of
which one is homeomorphic to a disc and the other to a Moebius strip.
The first component is called the interior of the oval. Therefore it makes
sense to talk of one oval lying inside another or including it. The problem
is to clarify the possible dispositions of ovals (in the sense of one being
included in another) for all real plane smooth curves of a given degree.
At present the answer is known for curves of degree 2, 4, or 6; we shall
describe the result in the case of maximal number of ovals. Using
the formula for the genus of a smooth plane curve of degree 2n we see
that the maximum number of ovals admissible by Harnack's theorem is
2n 2 - 3n+2.
A curve of degree 2 can consist of a single oval. A curve of degree 4
has not more than 4 ovals. Here only one disposition is possible: all 4
ovals lie outside each other. A curve of degree 6 has not more than eleven
ovals.
In this case three types of disposition are possible: one oval contains
inside 1, 5, or 9 ovals not including each other, and outside it there
lie 9, 5, or 1 oval, respectively, none including another.
Also certain general inequalities and congruences are known to which
the numbers of dispositions of ovals of one kind or another are subject.
For example, from one general result of Petrovskii it follows that the
number of ovals of a curve of degree 2n not containing each other does
not exceed !n(n - 1) + 1, from which it follows, in particular, that a curve
of degree 6 cannot split into eleven ovals not contained in one another.
A classification of all possible types of disposition of ovals is not known
at present.
In the same lecture and in connection with the same problem Hilbert
points to an analogy between the problem on ovals of a real algebraic
curve and the limit cycles of a differential equation
dy f(x,y)
--= ,
dx g(x,y)
where f and g are polynomials. In this problem not even an analogue to

Harnack's theorem has been found, that is, no estimate for the number of
limit cycles for such a differential equation in terms of the maximum
degree of the polynomials f and g is known. No such estimate is known
even when the maximum is equal to 2.
Exercises
1. Let k be an algebraically closed field of characteristic 0, k {t} the field of fractions
of the ring of power series in one variable. Show that this field has a unique extension of a
given degree n, namely that which is obtained by adjoining"yt. Hint: Use the arguments
in the proof of Theorem 1.
2. Let X and Y be smooth projective curves, f : X -> Y a morphism, and f(K) = Y.
Derive a formula that expresses the genus of X in terms of the genus of Y and the multiplicity
of the ramification points of the morphism f. Hint: Consider triangulations of X and Y
that are compatible with f.
3. Let X be a projective smooth model of the curve with the equation
y2 = (x- a) (x- b) (~- c) (x- d),
and f : X(CC)-> Pl(CC) the continuous mapping that corresponds to the morphism specifying
the function x. We denote by IX and fJ disjoint segments on the sphere p 1(CC) joining a to b
and c to d. Show that f-l(Pl(CC)-IX-P) splits into two connected components Xi each
of which is mapped by f homeomorphically onto Pl(CC)-IX- p, and that Pl(<C)-IX_ P
is homeomorphic to the sphere with two discs removed. - -
4. In the notation of Exercise 3, show that the boundary of the closure Xi is homeo-
morphic to (Xu p, i = 1,2, that X is obtained by identifying these boundaries and thus is
homeomorphic to a torus in accordance with Theorem 5.
5. Show that if a real curve of degree 4 splits into three ovals, then none of them lies
inside another. Hint: Otherwise there would exist a line intersecting the curve in six points.
6. Show that the normalization of a real curve is a real curve.
7. Consider the normalization of the projective closure of the curve
y2 = _ (x - etl ... (x - e2g +2)' ei # ej , ei E IR.
Show that the number of ovals for it is the same as the bound given by Harnack's theorem.
Chapter VID. Complex Analytic Manifolds
§ 1. Defmitions and Examples

1. Defmition. In the preceding chapter we have investigated the
topological space X(<C) connected with an arbitrary algebraic manifold X
defined over the field of complex numbers <C. The example of smooth
projective curves shows to what extent this space characterizes the
variety X. We have shown that in this case the only invariant of X(<C) is
the genus of X. We can say, therefore, that the genus is the only topological
invariant of a projective curve. Undoubtedly, the genus is a most im-
portant invariant of an algebraic curve, however, it does not determine
the curve by any means. We have seen in Ch. III, § 5.6, that there exist
very many non-isomorphic curves of one and the same genus. The
connection between a variety X and the space X(<C) for varieties of higher
dimensions bears a similar character.
When we focus our attention on the way in which the topology was
defined in X(<C) (Ch. VII, § 1.1), we observe that in the same manner we
can connect with X another object, which reflects vastly more properties
of this variety. We do this here under the assumption that X is a smooth
variety; the general case will be treated in § 1.5.
We beginju~t as we have done in the preceding chapter: we consider
a point x E X(<C), some system oflocal parameters t 1 , ••• , tn at this point,
and a homeomorphism determined by them
(1)
of some neighbourhood U of x and a neighbourhood of zero V C <cn • We

have used this homeomorphism to define on X(<C) the structure of a
topological 2n-dimensional manifold. An essential feature here is the
compatibility of the various mappings (1) that are defined in various
neighbourhoods U and by means of various systems of parameters. It
follows from the fact that if a function f E <C(X) is regular at x, then
g = f cp - 1 is an analytic function of n complex coordinates Z l' .•• , Zn in a
neighbourhood of zero in <cn • Of this property we have only used a very
344 Chapter VIII. Complex Analytic Manifolds
small part: the continuity of g, from which we have derived that any other
local parameters Ul' •.. , Un are continuous functions of t 1 , ... , tn.
At the basis of this argument lies the fact that the concept of a cort-
tinuous complex-valued (or real-valued) function in a neighbourhood V
of x E X(<C) can be defined in an invariant manner. It is natural to use
this name for a function h: U -+<C for which h· cp -1 is continuous in
V C <cn, and this property does not depend on the choice of cpo We now
recall that if f is a regular function at x, then g = f cp - 1 is not only con-
tinuous but analytic. Hence it follows that if U 1 , ... , Un is another system
of local parameters at x, then in some neighbourhood U' C U of this
point t 1, ..• , tn are analytic functions of Ul' ... , Un' Therefore, if for some
continuous function h: U -+<C the function h cp -1 is analytic in a neigh-
bourhood of zero, then the same property holds for every mapping (1)
given by another system of local parameters at x.
Thus, the following concept is well-defined.
A complex-valued function h defined in some neighbourhood of a
point x E X(<C) is said to be analytic at x if g(Zl' ... , zn) = hcp-l, defined by
means of(1), is an analytic function of the variables Zl' ... , Zn in a neigh-
bourhood of zero in- <cn •
The functions that are analytic at all points of an open set U form a
ring, which is denoted by (9ao(U), Since the defmition of being analytic
has local character, the mapping U -+(9ao(U) determines a sheaf (9ao' a
so-called sheaf of analytic functions. Evidently it is a subsheaf of the
sheaf of continuous function on X(<C), and on the other hand, the sheaf of
regular functions (9 is a subsheaf of it.
In preceding parts of the book we have determined an algebraic
variety by its topological space (in the spectral topology) and the sheaf
of regular functions. Similarly a topological space with a sheaf of analytic
functions given on it leads to a new concept which we wish to define.
To begin with we consider a domain W in the space <cn of complex
variables. For every open set U C W the collection of all functions that
are analytic at all points of U form an algebra (9ao(U) over <C, and the
assignation U -+ (9ao(U) determines a sheaf of algebras over <C, which is a
subsheaf of the sheaf of continuous functions on W. We call (9ao the sheaf
of analytic functions on W.
These ringed spaces play in our theory the role of the simplest objects
analogous to that of affine schemes in the definition of the general concept
ofa scheme.
Defmition. A ringed space (X, (9x) consisting of a topological Haus-

dorff space X and a sheaf (9x of algebras over <C given on it, which is a
subsheaf of the sheaf Qf continuous complex-valued functions, is called a
complex analytic manifold if it satisfies the following condition: every
§ 1. Dermitions and Examples 345
point x E X has a neighbourhood U such that the ringed space defined

by the restriction of (!Jx to U is isomorphic to (W, (!JaJ, where W is a
domain in ([:" and (!Jan the sheaf of analytic functions on W.
Continuous functions that are sections of (!Jx over an open set U C X
are called analytic functions on U.
A mapping f : X --+ Y of two analytic manifolds is called holomorphic
if it is continuous and determines a morphism of the ringed spaces. The
latter is equivalent to the fact that the mapping f* carries analytic
functions into analytic functions.
If x is a point of an analytic manifold, U a neighbourhood of x, and
f : U --+ W C ([:n an isomorphism onto an open set in ([:", then the number n
is called the complex dimension at x. From the definition it follows that
as a topological space X is a 2n-dimensional manifold at x. Therefore
the complex dimension is one and the same for two points of a single
connected component of an analytic manifold. If X is connected, this
number is called its complex dimension.
Above we have constructed the sheaf (!Jan on any smooth algebraic
manifold X defined over the field of complex numbers. Evidently
(X([:), (!Jan) determines an analytic manifold, which we denote by X an •
Every morphism f : X --+ Y of algebraic manifolds determines a holo-
morphic mapping Xan --+ Yan of analytic manifolds, which we denote by
fan. The remaining part of this chapter is mainly devoted to a study of
connections between algebraic varieties and the analytic manifolds
corresponding to them. For example, the following problem arises:
Is every analytic manifold of the form Xan , where X is some algebraic
variety?
Is every homomorphic mapping Xan --+ Yan of the form fan, where
f : X --+ Y is a regular mapping of algebraic varieties?
Does an isomorphism of analytic manifolds Xan and Yan imply an
isomorphism of the algebraic varieties X and Y?
The answers to all these questions are negative, and relevant examples
are not hard to construct. On the first two, see Exercises 1 and 3; on the
third see § 3.2. But if we restrict ourselves to compact manifolds, then
the same questions become much deeper and the answers to them less
trivial. We consider them in the next few sections.
2. Factor Spaces. Here we describe a new construction of analytic

manifolds. As a first application we obtain a number of examples, which
enable us to answer some of the problems discussed at the end of the
preceding subsection.
Let X be a topological space and G a group consisting of homeo-
morphisms of this space. We say that G acts on X freely and discretely if
the following two conditions hold: 1) every point x E X has a neighbour-
hood U such that

gUnU=0 (1)
for every g E G, other than the identity transformation, and 2) any two
points x, y E X for which y #- g x for any g in G have neighbourhoods U
of x and V of y such that
gUnV=0
for all g E G.
We denote by X/G the set of equivalence classes of points, where two
points Xl and Xz are called equivalent if there exists agE G such that
gXl = xz. By assigning to every point the class containing it we determine
a mapping
n:X-+X/G.
In X/G we introduce a topology, by calling a subset U C X/G open if
n-l(U) is open in X. Condition 2) guarantees that if X is a Hausdorff
space, then so is X/G. Ifx E X,y= n(x), U a neighbourhood ofx satisfying
(1), then V = n(U) is a neighbourhood of y. By virtue of (1)
n-l(V)=UgU, glUng z U=0 for gl#-gz, (2)
and the mapping
n:gU-+V
is a homeomorphism. A mapping of topological spaces n: X -+ Y having
the analogous property is called an unramified covering.
Now we assume that (X, (Dx) is an analytic manifold and that G, acting
freely and discretely on X, consists of automorphisms of this analytic
manifold. In that case we define on X/G a sheaf (DXIG' taking for (Dx1G(V)
the collection of those continuous functions j on V for which
n* jE(Dx(n-l(V)).
Let us show that (X/G, (Dx1G) is an analytic manifold. For this purpose
we consider a neighbourhood U of x E X that satisfies simultaneously
the condition in the definition of an analytic manifold and (1), and let
<p:U-+W
be an isomorphism of it with a domain in (Cn.
We set n(U) = V and consider a continuous function f on V. By (2)
j E (Dx/G(V) if and only if n* j E (Dx(gU) for all g E G. On the other hand,
since g is an automorphism of the analytic manifold X, it determines an
isomorphism of U and g U under which the restrictions of n to U and
gU go over into each other. Hence it follows that if
nl : U -+ V
is the restriction of n to U, then j belongs to (DXIG(V) if and only if ni j
belongs to (Dx(U). Since nl is a homeomorphism, it follows that n l is an
isomorphism of the analytic manifolds U and V. Therefore, we have the

isomorphism
qnr."1 1 : V --+ W ([n ,
whose existence shows that X/G is an analytic manifold.

Example 1. We regard the n-dimensional vector space ([n over the
field of complex numbers as a 2n-dimensional space over IR. and we choose
in it m linearly independent vectors (over IR) ai' ... , am. The set of vectors
of the form
a = Ii a 1 + ... + Imam' Ij E'll ,
is called an m-dimensionallattice, which we denote by D. The transforma-
tions
giz) =Z +a, Z E([n, aE D,
are automorphisms of the analytic manifold ([n. Since
all the ga' a ED, form a group G.Clearly, it acts on ([n freely and discretely.
For let us supplement ai' ... , am to a basis ai' ... , a2n and denote by U
the open set consisting of the vectors of the form
a=xl al + ... +X2n a2",X jEIR,
-!<Xj<! for i= 1, ... ,m.
Then the set z + U consisting of the vectors z + U, U E U, is a neighbour-
hood ofz and
for a E D, a ¥= O.
Thus, ([n/G is an analytic manifold, which is easily seen to be compact
if and only if m = 2n.
Suppose then that m = 2n. In that case ([n/G has a very ~imple topo-
logical structure. Since
([n = IRa 1 + ... + IRa 2", D ='llal + ... +'lla2n'
we see that ([n/G is homeomorphic to the product of 2n copies ofIR/T,
where T consists ofthe translations t --+ t + n, t E IR. n E'll. Evidently R/T is
homeomorphic to a circle, and ([n/G homeomorphic to a 2n-dimensional
torus. The varieties ([n/G (with m = 2n) are therefore called complex tori.
Later we shall see that as analytic manifolds they are by no means always
isomorphic.
Example 2. Let X = ([n - 0, c a positive number, c =f= 1, and G the
group consisting of the transformations
(Zl' ... ,zn)--+(dz 1 , .•• , C1zn), IE'll.
The fact that G acts on X freely and discretely can easily be verified
directly, but the subsequent arguments make it perfectly obvious.
We write any point Z E X in the form
z=r·u,
where r is a positive number and u = (u 1, ... , un) is such that
lu 1 12 + ... + IUn l2 = 1.
This representation is obviously unique and determines a homeo-
morphism of X onto
where IR+ is the set of positive real numbers and s2n-1 the sphere of
dimension 2n - 1. In this representation the transformations in G act
trivially on s2n- 1, and on IR+ as multiplication by a power of c. If we
map IR+ onto IR by means of the log-function, the latter action turns into
the translation by the lattice vector 7l10gc. Hence it is clear that G acts
freely and discretely and thatX/G is homeomorphic to (IR/Z log c) X s2n-1,
that is, to S1 X S2n-1. The compact analytic manifold we have constructed
is called a H opf manifold.
3. Commutative Algebraic Groups as Factor Spaces. We return to

Example 1 of the preceding subsection: the factor space ern/G, where G
consists of translations by vectors of a certain lattice Q. This lattice is a
subgroup of ern, and the factor space ern/G is homeomorphic to the factor
group ern/Q, therefore is a group. It is very easy to check that the mapping
m: ern/Q x ern/Q --+ern/Q ,
defined by the group law is holomorphic. Thus, ernjQ is a commutative
complex analytic Lie group.
Let us assume that the manifold ern/Q originates in some algebraic
variety X, in other words, is of the form Xan• In that case it can be shown
that m = /lan' where
/l:XxX--+X
is the morphism that defines in this way on X the structure of an algebraic
group. In the most interesting case when X is compact this follows from a
theorem of the next section. Thus, in this case X is an Abelian variety.
We show presently that, conversely, every commutative algebraic
group over the field of complex numbers can be represented in the form
ern/Q, where Q is a lattice. To do this we need one auxiliary result.
Lemma. An invariant one-dimensional differential form on a com-
mutative algebraic group is closed.
Proof. Let q> be an invariant differential form on a group G. Then dq>,

as is easy to verify, is also invariant. Therefore we need only show that
(dq> )(e) = 0, from which it follows that dq> = O.
Writing q> in the form
q> = !:.lpmdum,
we make use of formula (5) in Ch. III, § 5.2:
L Cmllpl = lpm(e) E <C .
I
From this it follows that
"L...,-::)-Cml+
011'1 "OCml
L...,1p1-::)-=0.
I vUj I vUj
We consider this equality at e. Since cmz(e)= <5m b it follows from this
equality that '
01pm (
-::)- e)+" OCml (e) =0.
L...,lpl(e) -::)-
VUj I VUj
To prove the equality
( 01pm) (e) = (Olpj) (e), (1)

OUj 0l!-m
which expresses the fact that q> is closed, it is sufficient for us to verify that
OCml
::) (e)= OCjl
::) (e.
)
VUj vUm
It is here that we use the commutativity of G. By the formulae (4) in
Ch. III, § 5.2:
Since the group is commutative,

Jl*(Um)(gl' g2) = L Vmj(gl)Wmj(g2) = LWmj(gl) VmJ{g2)'
j j
Therefore,
and consequently,
OCml (e) = L OVl,j (e) oWIj (e) = L oWIj (e) OVl,j (e) = OCjl (e).
oUj j oU j OUm j OUj OUm OUm
This proves formula (1) and the lemma.
Now we consider an arbitrary n-dimensional commutative algebraic

group A defined over the field of complex numbers. By the proposition in
Ch. III, § 5.2, the space of invariant one-dimensional differential forms
on A is n-dimensional. We denote by (01' ••• , (On a basis of it. According
to the lemma, the differential forms (Oi are closed, hence there exist
holomorphic functions fl' ... , fn defined in some complex neighbourhood
U of the zero element 0 E A such that
From the invariance of the forms (Oi it follows that

dt;/;=dfi
in the domain U (") t; U. In other words,
t; fi = fi + IX;, IXi E (C • (2)
But (t; fi)(gl)= /;(g + gl) (we write the group law on A additively). There-
fore (2) indicates that /;(g + gl) = fi(g) + IXi if g, g + gl E U. In particular,
setting g = 0 we see that
/;(g + gl) = /;(g) + /;(gd (3)
for g, g + gl E U.
Thus, the /; determine a "local homomorphism" of a neighbourhood
of 0 in A into a neighbourhood of 0 in (Cn. From the proposition in Ch. III,
§ 5.2, we see that the dofi form a basis in e:, and this means that the
Jacobian det( ~~ (e)) #0 for every system of local parameters t 1 , ••• , tn
at o. By virtue of this, the mapping cp
is an analytic isomorphism between the neighbourhood U of 0 and the

neighbourhood V of 0 in (Cn, and by (3) a "local isomorphism" of the
groups.
N ow we construct a homomorphism tp: (Cn _ A, by setting for Z E (Cn
tp(Z) = lcp-l (-r)

where the integer I is chosen sufficiently large so that z/l E V. That this
is well-defined (independent of l) follows at once from (3). So we have
constructed the homomorphism
tp :(Cn_A,
which on V C (Cn is the same as cp - 1. Hence it is easy to deduce that tp is

holomorphic. From the fact that it maps isomorphically onto V it follows
that 1p(CCn) = A. We denote by 0 the kernel of 1p. Then 0 n V = 0, that is,

o is a discrete subgroup ofCCn • By a standard argument it follows that 0
is a lattice. Thus, A ~ CCn/0, where 0 is a lattice.
4. Examples of Compact Analytic Manifolds that are not Isomorphic

to Algebraic Varieties. Here we are concerned with the first of the
problems stated in § 1.2 in connection with the concept of an analytic
manifold: does every analytic manifold originate in some algebraic
variety, in other words, is it of the form Xan? Since the problem only
becomes really interesting when we restrict ourselves to compact analytic
manifolds, we make this assumption in the statement of the problem.
The fact that with this restriction the problem becomes far more
subtle is clear if only because for one-dimensional varieties the answer
is in the affirmative: every one-dimensional compact manifold is
isomorphic to one of the form Xan, where X is a smooth projective curve.
This proposition is called Riemann's existence theorem. We do not prove
it here-in every form a proof requires certain arguments of an analytic
character connected with the theory of harmonic functions. A proof can
be found, for example, in [31], Ch. 8.
It is all the more interesting that for dimensions greater than 1 the
answer to our problem is in the negative. Here we come across a familiar
phenomenon in algebraic geometry: many difficulties do not arise in the
one-dimensional case. In this subsection we give some examples of
compact manifolds that are not algebraic, and for simplicity we confine
ourselves to the two-dimensional case.
Owing to the fact that this problem is connected with the most central
concepts, we analyse two construction principles for such examples: in
Example 1 we use almost exclusively algebraic arguments, and in
Examples 2 and 3 more geometrical ones.
Example 1. Our manifold is a complex torus, that is, of the form CC2 /0,
where 0 is a four-dimensional lattice. If it were an algebraic variety,
then by what has been shown in § 1.3 it would be Abelian. We now
establish a property of Abelian varieties which turns out to fail in some
part of Q. This property is called Poincare's theorem on complete
reducibility and reads as follows.
Proposition. If A and B are Abelian varieties and q>: A --+ B an epi-
morphism, then there exists an Abelian variety C C A such that dim C
= dimB and q> : C --+ B is an epimorphism.
For brevity we assume that dim A = 2, dimB = 1, and that the ground
field is of characteristic O. We shall apply the proposition only with these
restrictions.
We consider a point a E A and the subvariety Y = cp-l(cp(a)) contain-

ing it. By the theorem on the dimension of fibres, Y is a curve. There exists
another irreducible curve X passing through a, but not contained in Y.
This follows because A is algebraic; it is sufficient to consider an affine
neighbourhood of a and to take for X the closure of a suitable hyperplane
section or a component of it. The morphism 1p : X - B, the restriction of cp,
evidently has finite degree and the fibres 1p - 1 (b) are finite.
r
For a divisor D on X, D = L lix;, we denote by SeD) the point

i= 1
11 Xl EB··· EB lrxr E A, where EB is the group operation on A, and we set
feb) = S(1p*(b)).
We claim that f is a morphism of B into A. To begin with we verify

that it is a rational morphism. Let () be a primitive element of the extension
k(X)/k(B),
k(X) = k(B) (()) .
Consequently, the coordinates tl of some point X E X have the form

F l (()), Fl E k(B)(T). If ()1, ... , ()n are all the conjugates of () over k(B), then
the points Xi with coordinates Fl(()i) have a common image in B:
1p(X;) = 1p(x) = b .
Obviously, the coordinates of feb) can be expressed in terms of the
coordinates of the Xi rationally and symmetrically, that is, they are
symmetric functions of the ()i and are therefore contained in k(B). This
shows that f is a rational morphism.
From the fact that A is complete and B is a smooth curve it follows
that f is a morphism, and from Theorem 4 in Ch. III, § 3, that it is a
homomorphism. We set C = feB). To prove the proposition we have to
verify that cp( C) # O. But by definition
cpC=cpfB=vB,
where v is the endomorphism of multiplication by n in B,'

v(b) = bEB .. · EBb.
Since the ground field is of characteristic 0, Kerv is finite (Exercise 2 to
Ch. III, § 3), vB = B # O. This proves the proposition.
Now we can complete the construction of the example. The idea is to
construct a complex torus for which this proposition does not hold.
Suppose that a lattice Q in ([:2 has a basis of four vectors
(1,0), (i, 0), (0,1), (a, (3).
It is easy to verify that they are independent over IR, provided that {3 is
not real. We set A = <cz/Q, B=<c 1 /Q', where Q' is the lattice with the
basis (0,1) and (rx,{3). The mapping (ZI'ZZ)--+zz induces, as is easy to
see, a holomorphic homomorphism cp: A --+ B. We assume that A is an
Abelian variety. From Riemann's existence theorem it follows that B is
an algebraic curve (we shall verify this directly in Ch. IX, § 2). As we shall
see a little later (Theorem 2 of §3), it follows that cp is a morphism. Now
we may apply Poincare's theorem on complete reducibility and obtain
that there exists a one-dimensional Abelian variety C C A such that
cpC=B.
We denote by A the inverse image of C in <c z. This is a closed sub-
group of<C z, and all the closed subgroups in any IRn are easy to determine.
A simple argument (see [25J, Pontryagin) shows that they are of the form
7le 1 + ... + 7les + lRes+ 1 + ... + IRes+r> where el'···' es + r are independent
over IR. In our case A) Q, hence contains four independent vectors over
IR. We denote by Ao the connected component of zero in A. This is an
IR-admissible linear subspace. Since C is defined in A by a single local
equation and is smooth, Ao is defined in <c Z by a local equation
I(ZI' zz) = 0, in which the linear part does not vanish. Let
1=11 + Iz + ...
be the decomposition of the Taylor series for I into its homogeneous
constituents. Then for sufficiently small rx E IR
l(rxz 1 , rxZ Z ) = rxll +rx zIz + ... ,

and since Ao is a linear subspace over IR, we have Iz = 0, .... So we see that
Ao is given by the linear equation 11 = 0. As a result we obtain that
A=7le 1 +7le Z +IRe 3 +IRe 4 =7le l +7le z +Ao,
where Ao = IRe 3 + IRe 4 is a <C-linear subspace of <c z. In other words,

e4 = },e 3 , A. E <C. In conclusion we recall that A) Q. Hence it follows that
under the projection onto 7l e 1 + 7l ez a two-dimensional sub lattice of Q
goes into 0. Therefore AonQ is a two-dimensional sub lattice of Ao.
Consequently, we have reached the conclusion that Poincare's theorem
in our case simply indicates the existence of a complex line
for which A °n Q is a two-dimensional lattice and which projects onto the

whole line ZI' that is, does not coincide with the line Zz = 0.
In other words, to verify that the theorem holds we have to find in Q
°
a vector e for which Z 1 # and a vector A.e E Q for a non-real complex
number A.. Let us find out whether this is always possible. Let
e = a(1, 0) + b(i, 0) + c(O, 1) + d(a, P),

Ae = a (A., 0) + b(iA., O):l- c(O, A.) + d(A.a, A.P),
a, b, c, dE 7t .
All the z2-coordinates of the vectors in Q are contained in 7L + 7L p. In

particular, (c + dP)A E 7L + 7LP, therefore A. must be contained in the
field CQ(P) (we recall that c + dP ¥= 0, by hypothesis). Similarly, from a
discussion of the zl-coordinates we obtain that a E CQ(P, A., i) = CQ(P, i).
Evidently, this condition does not always hold: for example, we can set
p=i,a=t!2.
The construction of the example is now complete. It is interesting to
look once more at our arguments, in order to understand at what place
we have made essential use of the fact that A is an algebraic variety. It is
easy to convince ourselves that all the arguments except a single one can
be carried through equally for analytic manifolds. This only essential
argument occurs in the proof of Poincare's theorem, at the place where
we have drawn the curve X that does not coincide with ({J-l(b). From
this we can conclude that this property is not true for the torus constructed
in our example. Thus, this torus has a mapping ({J to a curve B such that
the only compact analytic manifolds in A are the fibres ({J - 1 (b). (More
details on the notion of a subvariety will be given in § 1.5, here we can
interpret this as the image of a projective curve under a holomorphic
mapping in A.) So we see that A is very barren in one-dimensional
subvarieties, and this is the main difference to algebraic surfaces, which
are loaded with curves all over.
Example 2. This also refers to a two-dimensional torus, but utilizes
some topological arguments. Again, let
Q=7Le l +7Le2+7Le3+7Le4'
Then A is homeomorphic to the torus (IRj71)4. Therefore H2(A,7L)=7L 6 ,
and their generators are the six cycles Si,j' 1:( i <j:( 4, the images of the
planes IRei + IRe j' i <j, in A.
In this example we begin with the same argument as at the end of the
preceding example. If A were algebraic, we could find in it an algebraic
cu~ve C C A. If v: CV -+ C is the normalization mapping, then by tri-
angulating C on the basis of Theorem 3 in Ch. VII, § 3, we would make C
into a singular cycle. In particular,
c'" L
1~i<j~4
ai,jSi,j' ai,j E 7L .
We show that the cycle C is not homologous to 0, consequently that

not all the aij vanish. To see this we observe that the differential form
1
2i (dz 1 1\ dZ 1 + dz z 1\ dzz ) on q::z is invariant under Q, therefore deter-
mines a differential form w on A. We show that Jw > 0, from which it
c
follows that C is not homologous to O. If we regard Zl and Zz as functions
in a neighbourhood ofa point XEC, then in a neighbourhood ofy=v-1(x)
our form is equal to
;i (ldV:~Zlr +ldV:~ZZr)dtl\dt>O, (1)
where t is a local parameter at y.

Now we consider in a similar way the differential form I] corresponding
to dZ 1 1\ dz 2 • On the one hand, by Stokes' theorem
JI] = Lai,j J 1],
C Si,j
where J I] is easy to compute: the reader is recommended to verify that

Si,j
if ei = (lXi' /3i), then

f I] =IXi/3j-IXj /3i'
Sid
On the other hand,
.f I] =0.
C
For by analogy with (1) v*(I]) is on C" equal to
dv*(z 1) dv*(zz) d 1\ d =
dt dt t t
°.
So we see that under the assumption that A is algebraic we have a relation
(2)
in which not all the ai,j vanish. Of course, it is not hard to choose IX 1, ... ,IX4'
/31'''' ,/34 so that the numbers IX i /3j' ... , IXj /3i are independent over 7L. The
corresponding torus is not algebraic.
It is easy to verify that this torus has even fewer one-dimensional
subvarieties than the one constructed in Example 1: in fact, it has no
compact one-dimensional analytic manifolds at all.
Notes. 1. We write the coordinates of the vectors of a basis of the
lattice Q in the form of a matrix of type
and consider the skew-symmetric 4 x 4 matrix A in which the elements

with i <j are the same as in (2). Then (2) can be written in the form of
ai,j
an equation
QAQ'=O, (3)
where Q' is the transposed matrix. The existence of an integral matrix A
satisfying this relation is evidently necessary for the torus corresponding
to the matrix Q to he projective or even algebraic.
Other conditions are given by inequalities of the type (1). To obtain
conditions as general as possible, we consider the form
1 - - - -
W= 2i(A 1 A1 dZ 1 /\ dZ 1 + A1 A2 dz 1 /\ dZ2 + A1 A2 dz 2 /\ dZ 1 + A2 A2 dz 2 /\ dz 2 ).
Arguing exactly as in the proof of (1) we see that JW;;;;' O. Furthermore,

c
it is easy to verify that if the torus <c 2 jQ is projective and C corresponds to
a hyperplane section in one of its embed dings, then w = 0 only for J
c
A1 = ,.1.2 = 0 (see Exercise 9). The last condition can be written differently.
Let Q* be the Hermitian conjugate matrix to Q. Then the 2 x 2 matrix
Q A Q* is Hermitian, as is easy to see, that is, it corresponds to a Hermitian
form F(x). A simple substitution shows that Jw = F(..1.), where A = (..1. 1 ,A 2 ).
c
Therefore the relation we have derived indicates that F is a positive-
definite form. We write this as follows:
QAQ*>O. (4)
Thus, the relations (3) and (4) are necessary for the torus corresponding
to the period matrix Q to be projective. Precisely the same relations
are necessary for an n-dimensional torus with the n x 2n period matrix Q
to be projective. They are called the Frobenius relations. It can be shown
that they are also sufficient for the torus to be projective. Some hint on
the idea of its proof is contained in Ch. IX, § 2.
2. In discussing the last example we could replace the formula (1) by
a reference to the proposition in Ch. VII, § 1.3. Namely, a word-for-word
repetition of the arguments given these shows that if Wcv is an orienting
cycle of the curve C, then v*(wcv) is not homologous to 0 on A(<C). The
reference to the triangulation of C can also be replaced by integration
over the cycle. This is a convenient way to proceed in the following
example.
Example 3. Let X be a Hopf manifold (Example 2 of § 1.2). Since X
is homeomorphic to Sl X s2n- 1, for n> 1 its two-dimensional Betti
number is equal to O. The proposition in Ch. VII, § 1.3, shows us that X
is not a projective variety. It is easy to prove that it is not algebraic.
5. Complex Spaces. Analytic manifolds are analytic analogues to

smooth algebraic varieties. To restrict ourselves only to this concept
would be very inconvenient: varieties with singular points can arise even
in the study of smooth algebraic varieties as subvarieties or images under
regular mappings. Furthermore, the majority of the arguments by which
we have attempted to show in Ch. V the necessity of introducing the
concept of a scheme is applicable in the analytic situation. The cor-
responding analytic concept is not used in the remainder of the book.
However, it would be a pity not to mention it at all. Therefore we give its
definition and indicate without proofs some of its basic properties.
We begin with a special case. Let W (<cn be a domain in the space of
n complex variables, and fl' ... , II functions holomorphic in W. We
denote by Y the set of common zeros of the functions fl' ... , II. We
define a sheaf (Dy on Y, by setting
where Vis an open set on Y, V = Y n V, Vopen in W (all open sets on Y

can be represented in this form), (Dw the sheaf of holomorphic functions
on W, and (fl' ... , II) the ideal generated by these functions. Since Y is
the set of common zeros of fv ... , II, the right-hand side does not depend
on the choice of the open set V. Topological spaces Y with the so defined
sheaves are called local models.
Now we come to the global definitions. We define a complex ringed
space as a topological space X equipped with a sheaf (D, which is a sheaf of
algebras over <C. Any open set U ( X is itself a ringed space if it is equipped
with the restriction of (D to U.
Defmition. A complex analytic space is a complex ringed space
(X, (D) such that every point x E X has a neighbourhood U that is iso-
morphic, as a ringed space, to some local model.
As in the case of schemes, the stalks of the structure sheaf of an
analytic space are local rings. If they do not contain nilpotent elements,
then the space is said to be reduced. In this case the sheaf (D is a sub sheaf
of the sheaf of continuous functions on X, and on a local model Y the
stalk (Dy consists of those functions that are induced on Y by functions on
W holomorphic at y. In this case continuous functions f E (Dy are called
functions on Y holomorphic at y. Henceforth we only deal with reduced
analytic spaces without saying so. In this context morphisms of analytic
spaces are called holomorphic mappings.
Suppose that a closed subspace X' (X has the following property:
for every point x E X' there is a neighbourhood U (X and functions
fl' ... , II holomorphic in it such that X' coinc;ides with the collection of
their common zeros in U. We equip X' with the sheaf obtained by
restricting to X' the functions holomorphic on X. It is easy to verify that in

this way we arrive at an analytic space. It is called a subspace of X.
An analytic space X is said to be reducible if X = X' u X", where X'
and X" are subspaces of it, other than X. It is not hard to show that any
analytic space X is the union of a family of irreducible subspaces:
X=UX.. ,
and that only finitely many XIX pass through every point x E X. From
now on we only consider irreducible spaces.
A point x E X is called simple ifit has a neighbourhood isomorphic to
an analytic manifold. It can be shown that the set of simple points of an
irreducible analytic space X is connected, hence as a connected analytic
manifold has a well-defined dimension. This number is called the dimen-
·sion of X. Every proper subspace of X has smaller dimension. In
particular, it can be shown that the set of singular (that is, non-simple)
points is a subspace. In view of this an analytic space X is the union of
finitely many analytic varieties (not closed in X): the set of simple points,
the set of simple points of the subspace of singular points, etc.
Proofs of these properties can be found, for example, in [17], Ch. I-V.
Complex spaces are analytic analogues of algebraic varieties and even
of schemes, at least in the sense that with every scheme X of finite type
over the field of complex numbers we can associate a certain complex
space Xan (here again we consider schemes and spaces that are not
necessarily reduced). Let us describe the construction of the space Xan •
With a scheme X we associate the topological space if = Xred(<C) of
complex points of its reduced subscheme (in the complex topology).
An affine scheme X of finite type over <C is obviously a local model,
and the open set W is the whole of <cN in which X is contained. It is easy
to verify that this model does not depend on the embedding X -+ JAN. The
structure sheaf on this model is denoted by lDan .
U
If X is any scheme of finite type over <C and X = U<i) an affine
covering of it, then the sheaves lD~~ on U<i) just defined together determine
a single sheaf lDan on the space X. The pair (if, lDan ) is the complex space
Xan associated with X.
In § 1.4 we have been faced with questions on the connections between
the concepts of a complex manifold and a smooth algebraic variety. Of
course, similar problems arise in the context of complex spaces and their
connections with arbitrary algebraic varieties.
The only positive result we have stated in § 1.4: Riemann's existence
theorem, has an analogue in this general case. Namely, every compact
reduced one-dimensional complex space is isomorphic to an algebraic
curve. This result can be derived from Riemann's existence theorem by
§ 1. Definiti{)ns and Examples 359
means of the normalization process of a complex space, about which we

want to say a few words, omitting all proofs.
A reduced complex space is called normal if the local rings (!Jx of its
structure sheaf are integrally closed. Following very closely the arguments
we used in the case of algebraic varieties, we can .construct for every
reduced and irreducible complex space X its normalization, that is,
a normal space XV and a holomorphic mapping v: XV --+ X having the
propertie,s of Theorem 1 in Ch. VI, § 1. If X is compact, then so is XV.
A detailed account of all the arguments is contained, for example, in [1J,
447. In the case of one-dimensional spaces, with which we are concerned
henceforth, the position is somewhat simplified, and the reader can try
to provide these arguments himself by way of an exercise (by no means
trivial).
Let X be a compact reduced one-dimensional complex space with
the structure sheaf (!J. According to Riemann's theorem XV is a projective
algebraic curve. On X we introduce the spectral topology in which the
finite subsets and X itself are closed, and we define the sheaf @ by the
property @(U)= (!J(U)n<C(XV). It is not hard to verify that in this way we
define an algebraic curve X, and that Xan = X.
Exercises
1. Construct an example of a holomorphic mapping g: ([I -->([1 that is not of the form
Ian' where I is a morphism I :!AI-->/A I .
2. Let X be a smooth irreducible curve, and I a holomorphic function on Xan . Show
that if I is bounded on the set X([), then IE ([.
3. Show that the disc Izl < j in ([I is not isomorphic to Xan for any smooth curve X.
4. Show that if A is an elliptic curve, then the number of solutions of the equation
mx = 0, x E A, m > 0 an integer, 0 the zero point on A, is equal to m2 . If A is an n-dimensional
Abelian variety, then the number of solutions of this equation is equal to m2n. Deduce
that a plane smooth cubic curve has nine points of inflexion (see Exercise 8 to Ch. III, § 2).
5. Show that a one-dimensional Hopf manifold is isomorphic to a complex torus.
6. Let X = ([2 - o)/G be a two-dimensional Hopf manifold. Show that the mapping
([2 -0-->!p 1 :(ZI' Z2)-->(ZI :Z2) determines a holomorphic mapping X -->!p1 whose fibres
are one-dimensional complex tori. .
7. In the notation of Exercise 6, show that on X there are no one-dimensional complex
analytic sub manifolds exept the fibres of a mapping X -->!p I.
8. Let X be the complex space ([2, g the automorphism g(zl' Z2) = (- ZI' - Z2)' G= {j,g}
a group of order 2. Show that the factor space X/G (see Exercise 1 to Ch. V, § 3) is a
complex space and isomorphic to the cone in JA3 with the equation x y = Z2.
9. Let X = ([2/Q be a complex torus,
1 - -
w = 2i (1,1112 dZ I II dZ I + Al ,12 dZ I II dZ 2 + Al ,12 dZ I II dZ 2 + 1)'21 2 dZ 2 II dz2 ),
C an analytic curve on X, x E C. Identify the tangent plane to X at x with ([2 by means of
the mapping ([2 --> X and introduce coordinates in this way. Show that if x is a simple point
on C and if the coordinates of the tangent vector to C at x are 11-10 11-2' and if Al iii + 2211-2 '" 0,
J
then w > 0 (that is, '" 0). Deduce that Sw > 0 if X is a projective torus and C the class
c c
of a hyperplane section.
§ 2. Divisors and Meromorphic Functions

1. Divisors. Now we return to the theory of analytic manifolds. The
problem we consider is to construct for them an analogue to the theory
of divisors. We must begin with an account of some simple properties
of the stalk (!) x of the structure sheaf of an analytic manifold. By definition,
the ring (!)x is isomorphic to the ring <C{Zl, ... , zn} of power series in
Z 1' ... , zn that are convergent in some neighbourhood of x (the neighbour-
hood depending on the series). This ring is very similar to the ring of
formal power series. In particular, it is a regular local ring, and the
analogue to Weierstrass' preparation theorem holds for it, which is
stated exactly and proved almost exactly as for formal power series.
A proof can be found in [30], Ch. I, § 2. From this theorem it follows word-
for-word as for formal power series that in the ring <C{Zl, ... , zn} de-
composition into prime factors is possible and unique. In particular,
this ring has no divisors of zero.
Let U be a connected analytic manifold and (!)(U) the ring of functions
holomorphic on the whole of U. This ring has no divisors of zero. For if
f, g E (!)(U) and f g = 0, then the set of points where f =I- 0 is open, and g = 0
on this set. But then g = 0 on the whole of U, by the uniqueness property
of analytic functions. The elements of the field of fractions of (!)(U) are
called meromorphic fractions on U. If V C U is a connected open subset,
then the restriction (!)(U)--+(!)(V) extends to an isomorphic embedding
of the field of merom orphic fractions on U into the corresponding field
on V. Frequently we shall identify two corresponding meromorphic
fractions.
Defmition. A divisor on an analytic manifold X is a covering X = Ua. U
by connected open sets and a meromorphic fraction CPa. on every Ua.,
which must satisfy the condition: cp;;lcpp is holomorphic and does not
vanish on Ua.n Up.
Equality of divisors and their addition are defined exactly as for
locally principal divisors on algebraic varieties. A divisor is said to be
effective if all the meromorphic fractions CPa. are holomorphic in their
open sets.
Theorem 1. Every divisor is the difference of two effective divisors.
Lemma. If two functions f and g are holomorphic at a point x E <cn
and are relatively prime as elements of the ring (!)x=<C{Zl, ... ,Zn}, then
there exists a neighbourhood U of x such that f and g are holomorphic in U
and relatively prime as elements of every ring (!)y, y E u.
Proof of the Lemma. When we multiply f and g by invertible elements
of (!) x and apply Weierstrass' theorem, we can achieve that f and g
become polynomials in Zl with coefficients in <C{Z2, ... , zn} with the
highest coefficient 1. Since they are relatively prime, there exist

U,VE<C{ZI, ... ,Zn} such that
(1)
and this equations holds in some neighbourhood U of x. Suppose

that f and g have a common factor hE (!)y, Y E U. Then hlr, and again by
applying Weierstrass' theorem we see that h differs by an invertible
element of (!)y from an element hI E <C{Z2' ... ' zn}. But hll!. and since the
polynomial f in ZI over the ring<C{z2' ... ' zn} has the highest coefficient 1,
we see that hI is invertible in <C{Z2, ... , zn}. This proves the lemma.
Proof of the Theorem. We may assume that the divisor D is given by a
covering Ua. and a collection of meromorphic fractions CPa., such that
CPa. = fjga. in Ua.,
fa. and gao are holomorphic on Ua. and relatively prime at every point
Y E Ua.. Then it follows from the uniqueness of the decomposition into
prime factors in (!) ythat the fa. determine a divisor D' and the gao a divisor D",
both these divisors being effective and D = D' - D". This proves the
theorem.
Evidently every effective divisor D determines some complex subspace
of X, namely the one that is given by the equation CPa. = 0 in the open set
Ua.. It is called the support of D and is denoted by Supp D. If D = D' - D"
is a representation in the form of a difference of effective divisors, then
by definition Supp D = Supp D' u Supp D". Making use of the concept of
dimension of a complex space that was introduced at the end of the last
section we can state the following result.
Proposition. The support of a divisor is of codimension 1.
First of all, we have to specify this subspace by a more economical
system of equations. For this purpose we decompose at every point
x E Ua. the function CPa. into irreducible factors in (!) x' and we denote by
tpx the product of these factors to the first power. The function tpx is
holomorphic in some neighbourhood Ux of x, and all these functions
determine the same subspace Supp D as CPx (although possibly another
divisor). Thus, we may assume that from the very beginning the divisor
is given by functions CPx having no mUltiple factors in (!)x, x E Ua..
According to Weierstrass' preparation theorem we may assume that
for some point x E Ua. the function CPa. is given in the form
CPa. = z';+ aIz';-1 + ... +am ,
where ajE<C{z2, ... ,Zn}, and Zl' ... 'Zn are local parameters at x. By
virtue of the assumption about the functions CPa. made above we may take
it that 8cpj8z 1 is coprime to CPa. in (!)x, hence 8cpj8z 1 is not identically 0
on Supp D in a neighbourhood of x. Now we divide the points y e Supp D

into two types: those at which all the OepJOZi=O for i= 1, ... ,n, and
the remaining points. Evidently the first points form a subspace
S C Supp D, and as we have just seen, S:I: Supp D.
The proposition is an obvious consequence of two assertions:
a) the points of the first type are singular points of the subspace Supp D,
that is, in their neighbourhood this subspace is not isomorphic to an
analytic manifold, and b) in a neighbourhood of the points of the second
type Supp D is isomorphic to a manifold of dimension n - 1.
The assertion a) follows from the representation of the local ring of
ye SuppD:
(!)"SuPPD = (9 x/(epJ (2)
(the verification of (2) is left to the reader). If y is a point of the first

type, then ep.. em;, where mx is the maximal ideal of the local ring (9 x.
From this it follows at once that (!)"suppD is not a regular local ring,
hence that Supp D is not a manifold.
The assertion b) is a direct consequence of the implicit function
theorem. If, for example, OepJOZ1 (Y):I: 0, then Z1 is a holomorphic func-
tion of Z2' ... ' Zn on Supp D in a neighbourhood of y, hence Z2' ... ' Zn
determine an isomorphism of this neighbourhood with a domain in <en - 1 .
We do not develop the theory of divisors on analytic manifolds any
further. It can be carried through to results completely analogous to
those we have obtained for algebraic varieties. Namely, every divisor has a
unique representation in the form of a linear combination of irreducible
effective divisors, and irreducible divisors correspond one to one to
complex subspaces of codimension 1. Proofs of these facts are contained
in the book [36], Appendix to the Russian edition, 185-202. They are
quite elementary and do not depend on other parts of that book.
2. Meromorpbic Functions. Now we consider meromorphic functions

on analytic manifolds, which are an analogue to rational functions on an
algebraic variety. The basic auxiliary tool is the concept of a meromorphic
fraction, which was introduced in § 2.1.
Definition. A meromorphic function on an analytic manifold X is given
by a covering X = U Va. by connected open sets and a system of mero-
morphic fractions ep.. on V.. such that the restrictions of ep.. and epp to the
set Va.n Vp are identical for any oc and {3. Such systems of functions
are called compatible.
A covering X =UV. and a compatible system of functions ep ..
determine the same meromorphic function as the covering X = lip and U
system 'ljJp if the restrictions of ep .. and 'ljJp agree on V.. n lip.
If qJa, is a meromorphic fraction on Ua" qJa, = fig, f and g holomorphic

in Ua" and g(x) =1= 0 at some point x e Ua" then qJa, coincides with the
holomorphic function fig in a neighbourhood of x. This concept can be
carried over naturally to meromorphic functions. Thus, for every
meromorphic function qJ on X there exists an open set U C X and a
function f holomorphic in U such that the restriction of qJ to U coincides
with f We say that qJ is holomorphic at the points of U.
Algebraic operations on meromorphic functions are defined in terms
of the corresponding meromorphic fractions. Evidently all meromorphic
functions on a manifold X form a ring. If X is connected, then the ring
of meromorphic functions on it is a field. For let qJ be given by a covering
{Ua,} and a compatible system of functions qJa,. If qJ =1= 0, then at least one
qJa, =1= O. But from the compatibility of these functions it follows that then
qJp =1= 0 for all f3 for which Ua,n Up is not empty. Since X is connected, it
then follows that all the qJy =1= 0, hence the function qJ -1 given by the
system qJ:; 1 exists. Henceforth we only consider connected manifolds.
The field of meromorphic functions on such a manifold is denoted
by V#(X).
If the manifold is of the form Xan, where X is an irreducible smooth
algebraic variety, then clearly the rational functions on X determine
meromorphic functions on Xan. In other words, CC(X) C V#(Xan). Of
course, in general, equality does not hold. However, if X is complete,
then the two fields are the same, as we shall show in § 3.
A comparison of the definitions of the two concepts shows that every
meromorphic function qJ determines a divisor, which we denote by (qJ).
From the definition it follows that (qJ) is an effective divisor if and only
if qJ is holomorphic on the whole manifold X. For compact connected
manifolds this is possible only if qJ is a constant, just as in Corollary 1
to Theorem 2 in Ch. I, § 5.
Theorem 2. A function qJ that is holomorphic at all points of a compact

connected manifold X is constant.
The function IqJl obviously is continuous on X and therefore attains a
maximum at some point xo' We consider a neighbourhood U of x o,
isomorphic to an open set vcCC n •
We may assume that V consists of the points (Zl'"'' zn), ~IZiI2 < 1,
and under the isomorphism f: U --* V we have f(xo) = 0 = (0, ... ,0). The
function tp = (f - 1)* (qJ) is holomorphic on V and its modulus attains a
maximum at O. For every point (oc 1 , ... ,ocn)e V we consider the one-
dimensional complex subspace Zi = exit, i = 1, ... , n. On it the function tp
determines a holomorphic function of a single argument t, which by the
maximum modulus principle is a constant. Hence it follows that tp is
constant on Vand therefore cp is constant on U. Since X is connected, by

the uniqueness theorem cp is a constant on the whole of X.
Since for divisors of meromorphic functions the following identical
relation holds:
the theorem has the following corollary.

Corollary. On a compact manifold a meromorphic function is uniquely
determined by its divisor, to within a constant factor.
Having introduced the concepts of meromorphic functions and their
divisors, we can throw new light on the examples worked out in § 1.4 of
compact analytic manifolds that are not algebraic varieties.
We begin with Example 2: the two-dimensional torus A, which is not
algebraic because there is not a single algebraic curve on it.
As was shown in § 1.5, one-dimensional complex subspaces are
algebraic curves. Therefore on the torus A there is not a single one-
dimensional complex subspace, that is, no divisor different from O.
Hence it follows that the divisor of every meromorphic function on A is 0,
hence that all these functions are constants, by virtue of Theorem 2.
In other words, Jt(A) = <C. So we have a new characterization of the
non-algebraic torus A: on it there are far fewer meromorphic functions
than on an algebraic variety on which necessarily all rational functions
are meromorphic.
We now look at Example 1. There we have constructed a two-
dimensional torus A and a homomorphism f: A --+ B of it onto an
elliptic curve. This torus is non-algebraic because the only irreducible
curves lying on it are the fibres f-1(b).
The divisor of an arbitrary meromorphic function cp on A can be
represented, on the basis of Theorem 1, in the form
(cp) = D' - D" ,
where D' and D" are effective divisors. From the proof of this theorem it
is easy to see that the set SuppD' nSuppD" may consist only of isolated
points, and since distinct fibres f - 1 (b) do not intersect, in general, we have
SuppD' = Uf- 1 (b;) , SuppD" = Uf- 1 (bj),
Applying this reasoning to the functions cp - c, C E <C, we verify that

every meromorphic function on A is constant on the fibres of f. At a point
a E A we choose local parameters z 1, Z2 so that z 1 = f*(t) and t is a local
parameter at b = f(a). In this coordinate system cp can be represented as a
meromorphic fraction F(Zl' Z2) that does not depend on Zl' in other
words, is locally of the form f*(1p), where 1p is meromorphic on B. Hence

it follows that also on the whole of A the equality qJ = f*(1p), 1pE A(B),
holds. But B is an algebraic curve, and according to a theorem we shall
quote and prove in § 2.3, A(B) = <C(B). So we have shown that for
the torus A
A(A) = f*<C(B) .
We see that the non-algebraic character of A is again is expressed in
this relation. For an algebraic surface X the field A(X) contains <C(X)
and is therefore of transcendence degree at least 2 over <C, whereas in
our case the transcendence degree is 1.
The same arguments are applicable to Example 3 in § 1.4 (see Exer-
cises 6 and 7 to § 1). On a Hopfsurface X the equality A (X) = <C(lP1) holds.
3. Siegel's Theorem. The examples at the end of the preceding

subsection show that on a compact analytic manifold X there can be
"few" meromorphic functions compared with an algebraic variety of the
same dimension: more precisely, the transcendence degree of A(X)
can be less than the dimension of X. Quite a number of important
properties of compact analytic manifolds follow from the fact that there
cannot be too "many" merom orphic functions on them. This is what
we are going to prove next.
Theorem 3. The transcendence degree of the field of meromorphic
functions on a compact analytic manifold does not exceed the dimension of
the manifold.
The proof of this theorem is completely elementary. We preced it by
a simple remark.
Schwarz's Lemma. Let f(z 1, ... , Zn) = f(z) be a holomorphic function
in the domain IZil ,;;; 1, i = 1, ... , n, and M = max If(z)l. If f E m~, where mo
Iz;j", 1
is the maximal ideal of the local ring of functions analytic at the origin
of coordinates (that is, if all the derivatives of f of order ,;;; h vanish at
the origin of coordinates), then
If(z)I<Mmaxlzilh for Iz;l<1, i=1, ... ,n. (1)

i
Proof For a point z=(z1, ... ,Zn)E<cn we set Izl=m;lXlzJ

, For
a fixed z with Izl < 1 we set g(t) = f(tz), t E <C. This function is holomorphic
for It I ,;;; Izl- 1, and the first h coefficients of its Taylor series vanish at o.
Therefore g(t)/th is holomorphic for It I ,;;; Izl- 1 . By the maximum modulus
principle, in this closed disc we have Ig(t)/thl,;;; M/lzl- h= Mlzlh. Setting
t = 1 we obtain (1).
In the proof of the theorem we use the fact that the dimension of the
space of polynomials in v variables of degree .;;; / is given by
(/ + 1)(/ + 2) ... (1 + v)
v!
We denote this number by Hv(l). It is a polynomial in / of degree v.
Proof of the Theorem. Let fl"'" fn+l be n+ 1 meromorphic func-
tions on an n-dimensional compact manifold X. Our aim is to construct
a polynomial F(T1, ... , T,,+1) for which
(2)
For every point x E X we choose three neighbourhoods Ux ) Vx ) vv.,.
The neighbourhood Ux is such that in it
(3)
f.
Ji = P;,x , i= 1, ... ,n+ 1 ,
Qi,x
with P;,x and Qi,x being holomorphic in Ux and relatively prime in every
point of this neighbourhood. The existence of such neighbourhoods
follows from the lemma in § 2.1. The neighbourhood Vx together with
its closure is contained in U x' In it there exists a local system of co-
ordinates (Z1,'''' zn) such that Izi < 1. The neighbourhood vv., is given
by the condition Izl < 1
From the fact that for distinct points x and y the expressions (3) give
representations of one and the same function .r;, and that Pi,x and Qi,x
are relatively prime, it follows that
Qi,x = Qi,y q>i,x,y ,
where q>i,x,y is holomorphic and non-zero in Uxn Uy.
From the system vv., we select a finite covering of X (here we use that
it is compact):
X=U~·
We denote the number of sets ~ (and of points ~ E X) by r and we set
n+1
q>~,~ = TI q>i,~,~, C = max max 1q>~,~I·
i=1 ~", V"nV.
Observe that 1q>~,~1 is bounded in ~n v", because the closure of this set
is contained in U~n U~, where q>~", is holomorphic. Furthermore, C> 1
since q>~,~q>~,~ = 1.
For the polynomial F(T1, ... , T,,+1) of degree / in T1, ... , T,,+1, which
is not yet defined, we set
where
n+1
Qx= Il Qi,x'
i=l
Clearly R~ = q{'1 R" in ~n v;,.

Having introduced the notation we can now go over to the business
part of the proof. As a first approximation to (2) we show that for every
given h the polynomial F can be chosen so that F =ft 0 and
R~Em~ (4)
for all r points ~.
These conditions can be written in the form of relations
(DS R~)(~) =0,
where D S is a partial derivative of order < h. Therefore, there are linear
relations on the coefficients of F. The number of relations is rHn(h -1).
If we choose the degree I of F so that
(5)
then we can find a non-zero polynomial F for which (4) holds.
By Schwarz's lemma, with this choice of F the functions R~ are small
in the neighbourhoods W;: if
M = max max IR~(x)1 ,
~ XEV"
then
M
IR~(x)l.:;;: 2!' for x E W; . (6)
This tells us that M = 0, in other words, that (2) holds, for sufficiently
large I and h. For suppose that the maximum M is attained at a point
Xo E v;,. Then Xo E W; for some point ~. Therefore
M = IR'1(x o)1 = IR~(xo)II<p~jxoW.

If I and h are already such that (5) holds, then (6) is also true, and hence
M C 1.
M.:;;:2!'
It remains for us to choose I and h so that apart from (5) we also have
C112h< 1,
and then we find that M = O. The required choice is possible: if C = 2'\

A> 0 (because C> 1), then it is sufficient to take I < A-I h and satisfying (5).
For example, for I = him, where is m is any integer greater than A, we
obtain in (5) on the left-hand side a polynomial of much higher degree
in h than on the right-hand side, hence the left-hand side for sufficiently
large h divisible by m is, in fact, larger than the right-hand side. The
proof of the theorem is now complete.
By means of similar arguments it can be shown that if the trans-
cendence degree of the field vIt(X) is I and if fl' ... ' h are algebraically
independent meromorphic functions on X, then the degree of an ir-
reducible relation
F(J, fl'···' fz)= 0,
satisfied by an arbitrary meromorphic function f is bounded above.
Therefore the field vIt(X) not only has finite transcendence degree, but
is even finitely generated.
Exercises
1. Define an analytic vector bundle by analogy to what we have done in Ch. VI, § 1.2,
with this difference that E and X are analytic manifolds and p: E ..... X a holomorphic
mapping. Show that the correspondence between bundles and transition matrices that
was established in Ch. VI, § 1.2, remains valid for analytic vector bundles.
2. Show that the association of a divisor with a linear bundle, as described in Ch. VI,
§ 1.4, carries over to analytic bundles. Here we have to state the definition of this association
in Ch. VI, § 1.4, in terms of transition matrices [Formula (2) of Ch. VI, § 1.4]. Show that
also for analytic bundles equivalent divisors determine isomorphic bundles.
3. Let X be an analytic manifold, U. C X an open set for which there exists an iso-
morphic mapping onto an open set in CCn, and z~·), ... , z~·), the inverse images of coordinates
in CCn relative to this isomorphism. If Up is another such open set, then in U. n Up we set
8(zr), ... , z~'»
((I.p = "(z{p) z{p» .
u 1 " •• , n
Show that ((I.P are the transition functions of some linear bundle :%". Show that if X = Y".,
where Y is an algebraic variety, then:%" = K an, where K is the bundle corresponding to
the canonical class on Y. In the general case:%" is called the canonical bundle.
U
4. Let X be an analytic manifold, X = U. a covering for which there are isomorphisms
((ICC U...... CC· onto open sets of CC·. Suppose that in CCn functions f. are given that are holo-
morphic on ((I.(U.) such that under the isomorphisms ((Ip((l;; 1 : ((I.(U.n Up) ..... ((Ip(U.n Up)
the forms f. dz 1 /\ ... /\ dz. and fp dz 1 /\ ... /\ dZn are carried into each other. By definition
such a collection determines a holomorphic form w on X. Show that the functions ((I:(f.)
determine a divisor on X, the so-called divisor of the form w. Show that the divisors of
any two holomorphic forms are equivalent. Show that if a holomorphic differential form
exists on X, then the linear bundle defined by its divisor is isomorphic to the canonical
bundle.
S. Show that the canonical bundle of a complex torus is trivial.
6. Let Q be a 2n-dimensional lattice in CCn, X =CC·/Q an n-dimensional torus (Ex-
ample 1 of § 1.2), and x: Q ..... CC* a homomorphism of the group Q into the multiplicative
group of non-zero complex numbers. Define an action of Q on the space CCn X CC ' by the
condition
a(x, z) =(x +a, x(a) z),xeCC·,zeCC ' , ae Q.
Show that Q acts on CC· x CC ' freely and discretely. The projection CCn X CC ' ..... CCnis permutable
with the action of Q and determines a mapping p: (CC· x CC')/Q ..... CCn/Q = X. Show that p is
holomorphic and determines in Ex = (q;n X q;)/Q the structure of a linear bundle over X
(see Exercise 1).
7. In the notation of Exercise 6, show that two bundles Ex and Ex' are isomorphic if
and only if there exists a holomorphic function g on q;n, vanishing nowhere, such that
g(x + a) g(X)-l = x'(a) x(a)-l for all x Eq;n, a E Q.
8. In the notation of Exercises 6 and 7 assume, in addition, that Ix(a)1 = 1 for all a E Q.
Show that if X and X' have this property, then Ex and Ex' are isomorphic only if X = X'.
9. Show that Theorem 3 of Ch. VI, § 1 has no analogue in the theory of analytic
linear bundles: not every linear bundle is determined by some divisor.
§ 3. Algebraic Varieties and Analytic Manifolds

1. Comparison Theorem. Now we are in a position to prove some funda-
mental facts, which show that for complete and projective algebraic
varieties X over the field of complex numbers many properties of the
corresponding analytic manifold Xan can be reduced to algebraic
properties of X.
Theorem 1. If an algebraic variety X is complete, then a meromorphic
function on the analytic manifold Xan is a rational function on X.
Let f be a meromorphic function on Xan. According to Theorem 3
of § 2, it is algebraic over <C(X) (because Xan is compact). Therefore
we need only show that a function f that is meromorphic on Xan and
algebraic over <C(X) is rational on X. In the proof of this fact the com-
pleteness of X does not play any role.
Let
F(f)=fm+ad m- 1 + ... +am=O
be the irreducible equation over <C(X) whose root f is. By removing
from X the poles of the rational functions ai' we may assume that the ai are
regular on X. Then f is holomorphic on X. This follows from the fact that
in the rings @x.an the decomposition into prime factors is unique, so that
they are integrally closed in their fields of fractions.
In the product X x JA 1 we consider the set X' of points (x, z) satisfying
the relation
zm+ a1(x) zm-l + ... + am (x) =0.
The algebraic variety X' is irreducible, and <C(X') = <C(X) (f). We

denote by p : X' ~ X the natural projection. Once more we diminish X
and X', by removing from X the points at which the discriminant of the
polynomial F(T) vanishes, and from X' the inverse image (relative to p)
of this set. We denote the so obtained irreducible varieties by X and X',
as before. In this way we achieve that p - 1(x) for every point x E X
consists of m distinct points, and at every such point (x, z) we have
F~(x, z) =1= O.
If Zl' ... , Zn are local parameters at x E X, then it follows that

P*(Zl), ... ,p*(zn) are local parameters at any point p-1(X). Therefore
there exists a sufficiently small complex neighbourhood U of x such that
p-1(U) splits into m disjoint sets U1, ... , Um and the projection p: Uj --7N
is an isomorphism of the corresponding analytic manifolds. It is sufficient
for us to prove that p is a homeomorphism. This assertion means that
X'(<C) is an unramified covering of X(<C).
The function f determines a section of this unramified covering,
that is, a continuous mapping
qJ(x) = (x, f(x») ,
for which p.qJ = 1.
The information we have obtained is now enough to show that m = 1,
hence f E <C(X). For ifm > 1, then qJ(X) ¥= X; because qJ(x) is a single point,
whereas p-1(X) consists ofm points. We show that the sets qJ(X(<C») and
X'(<C) - qJ(X(<C») are closed and disjoint, from which it follows that X'(<C)
is disconnected. This contradicts the theorem in Ch. VII, § 2, because
X' is an irreducible algebraic variety.
All the assertions that remain to be verified are of local character,
so that we need only verify them for the sets U and p-1(U) instead of X
and X', where U is any neighbourhood of x E X. In particular, we may
take U to be connected and such that
Then qJ(U) must coincide with one of the Ub and all we need then clearly
follows.
Theorem 2. If X and Yare complete varieties, then any holomorphic
mapping f : X ao --7 Yao is of the form gao' where g : X --7 Y is a morphism.
Let x E X, Y = f(x), and U be an affine neighbourhood of y. We assume
that U C JAN and denote by t 1, ... , tN coordinates in this space. According
to Theorem 1 the f*(t j ) are rational functions on X.
If we can show that they are regular at x, then we have f = gao in
some neighbourhood of x. In this way we construct a system of morphisms
gj: lii--7 Y on open sets lii covering X. Clearly they determine a single
morphism g: X --7 Yfor which f = gao.
Thus, everything is reduced to a local proposition.
Lemma. If the rational function g is holomorphic at x, then it is regular
there.
We set g = u/v, u, v E (!)x. We denote by (!)x ao the ring of all functions
that are holomorphic in some neighbourhood of x, and by {jj x the ring of
formal power series. By assigning to a holomorphic function its power
series we define an embedding @x,an C @x such that

@xC@x,anC@x'
The fact that g is holomorphic at x means that vlu in @xan' Then a

fortiori vlu in @x. But according to 1 and 2 at the end of Ch.' II, § 3.3, it
then follows that vlu in @x, hence that g is regular at x.
Corollary. If for two complete algebraic varieties X and Y the analytic
manifolds Xan and Yan are isomorphic then X and Yare isomorphic.
Theorem 3. If X is a projective variety, then any compact analytic
submanifold Vof Xan is of the form Yan , where Y is a closed algebraic
subvariety of X.
Since X is contained in a projective space, it is sufficient to prove
the theorem for the case X = JPN. Furthermore, it is sufficient to prove
the theorem for connected submanifolds Vc JP,;;" because from the
compactness of V it follows that it consists of finitely many connected
components. Therefore, in what follows, we assume V to be connected.
We denote by Y the closure of V in the spectral topology ofJP N, that
is, the intersection of all algebraic subvarieties of this space containing V.
Let us show that the projective variety Y is irreducible. To see this we
need only show that the homogeneous ideal determined by it is prime,
that is, if P and Q are homogeneous polynomials such that p. Q = 0 on V,
then P = 0 or Q = 0 on V. If P does not vanish identically on V, then the set
U C V of those points where P¥-O is open in V. Suppose that the homo-
geneous coordinate Xo does not vanish identically on V. Then the function
QXol, 1= deg Q, vanishes on U and is holomorphic. By the uniqueness
theorem it vanishes on the whole connected component V containing U,
that is, on the whole of V. But this means that Q = 0 on V.
From the definition of Y it follows that every rational function
q> E <C( Y) determines a certain meromorphic function on V. In other words,
<C(Y) C /H(V). (1)

We set dim V = n, dim Y = m. Since in a neighbourhood of each of its
simple points Y is a 2m-dimensional manifold and Y) V, we have m> n.
But according to Theorem 3, § 2.3 the transcendence degree of the field
A1(V) does not exceed n, so that from (1) we obtain.
dim Y = dim V = n . (2)

From (2) it is easy to derive that Y..n = V, which is what we have to
show. For we denote by S the set of singular points of Y. The algebraic
variety Y - S is irreducible, hence by Theorem 2 of Ch. VI, § 1, the
manifold (Y - S)an is connected. The set V - (V (IS) is closed in (Y - S)an,
because is V is closed in Y..n. On the other hand, from (2) it follows that
v - (V nS) is open in (Y - S)an. Therefore V - (V nS) = (Y - S)an' that

is, (Y - S)an C V. Since V is closed and since by Lemma 1 of Ch. VII,
§ 2.1, (Y - S)an is everywhere dense in Yan> we have ¥an C V, that is, ¥an = V.
This proves the theorem.
2. An Example of Non-Isomorphic Algebraic Varieties that are

Isomorphic as Analytic Manifolds. We construct two algebraic varieties
X and Y that are not isomorphic, but for which the analytic manifolds
Xan and ¥an are isomorphic. According to the corollary to Theorem 1
in this situation X and Y cannot be complete.
First we give a description of the relevant example. Let C be a smooth
projective curve of degree 3, 0 a point of it, B the incomplete curve
C - 0, and p one of its points. In Ch. VI, § 1.4, we have seen that to every
divisor on B there corresponds a linear bundle E-+B. We take for X
the bundle corresponding to the divisor p, and for Y the direct product
B x /AI corresponding to the zero divisor. We have to prove two facts:
1) the algebraic varieties X and Yare non-isomorphic, and 2) the analytic
manifolds Xan and ¥an are isomorphic.
1) First of all we observe that X and Yare non-isomorphic as linear
bundles. According to Theorem 3 in Ch. VI, § 1, to prove this we need
only show that the divisors corresponding to them are non-equivalent,
that is, that p is not equivalent to zero on B. If this were so, then there
would exist a function f regular on B having a simple zero at p. The
divisor of this function on C must be of the form p - 10. By the corollary to
Theorem 1 in Ch. III, § 2, 1= 1, and by Theorem 1 at the same place this
contradicts the fact that C is a non-rational curve.
We now assume that there exists an isomorphism q>: X -+ Y of
algebraic varieties. We denote by Px and py the projections of X and Y
onto B defined by specifying the structure bundles on them. For every
point bEB the curve Pxl(b), hence also q>Pxl(b), is isomorphic to JAl .
If pyq>px l(b) is not a point, then py determines an embedding of <C(B) in
<C(q>Pxl(b)), which contradicts Liiroth's theorem, because the curve
q> Px 1 (b) is rational, but B is not. Thus, q> carries a fibre of X into a fibre of Y.
We see that there exists a mapping 1.p: B-+B such that the diagram
commutes. If Sx is the zero section of X, then 1.p = pyq>sx, from which it

follows that 1.p is a morphism, hence an automorphism of B. We denote by
1.p x 1 the automorphism of the bundle Y = B x JA 1 acting as 1.p on Band
trivially on AI. Then ep' = (1p x 1) -lep is also an isomorphism of X and Y,

but now pyep'pi 1 (b) = b for bE Band 1p' = pyep'sx = 1.
We set ep'sx = t : B --+ Y. This is a section of the bundle Y. We now
recall that Y is a vector bundle, so that it makes sense to talk of subtraction
of vectors in one ofits fibres. We set
ep"(x) = ep'(x) - tpx(x) .
Obviously, this is again an isomorphism of the varieties X and Y, however,
now not only does every fibre go into itself, but the zero point is preserved.
But the only automorphisms of the line A l preserving the zero point
are the linear transformations IX --+ A.IX. Thus, ep" must be an isomorphism
of the vector bundles X and Y, but as we have seen, they are non-iso-
morphic.
2) We make use of the fact that the association D--+LD in Ch. VI,
§ 1.4, carries over verbatim to analytic manifolds and meromorphic func-
tions (see Exercises 1 and 2 to § 2). In particular, if we can show that the
point p is the divisor of some meromorphic function, then this shows that
Xan and ¥an are isomorphic manifolds (even as "analytic vector bundles").
Thus, our task reduces to constructing a holomorphic function on B
having a single zero of order 1 at p.
To make everything quite specific, we assume that C is given by the
equation
(1)
and that 0 is the point at infinity on it. Then B is given by the equation (1)
in the affine plane.
We consider on C three rational differential forms:
dx dx 1 y-Yo dx
WI = y,W2=X y ,W3= 2 X-Xo ' y '
where p = (xo, Yo). Let us investigate their behaviour at o. A local para-

meter at this point is t = x/y, and
U v
X= 2'y= 3,U,VE((Jo,U(0)=v(0)= 1.
t t
Hence it follows that WI is a regular form at 0, W2 has there a pole of
order 2, and W3 a simple pole. From (1) (divided by y2) it is easy to derive
that xt 2 == 1(t4), yt 3 == 1(t4). Hence it follows that
W2 = (- ~ + f) dt, W3 = (- +g) + dt,j, gE ((J • (2)
Since the genus of C is 1, according to the results of Ch. VII, § 3.3,

the topological space C(d:) is homeomorphic to a torus. We denote by IX
and j3 a basis of its two-dimensional homology group, for example, its

parallel and meridian.
It is easy to see that W 1 is a regular form on C, and that W z has only
a single pole of order 2, namely at o. The integral of W l over a one-
dimensional cycle (J depends only on its homology class: if (J is
homologous to arx + bj3, then
SW l
tT
=as"
Wl +b SWl·
P
Although the form W z is not regular in the neighbourhood of 0, its integral
over a small contour around 0 is 0, because its expansion (2) has no term
with 1/t. Therefore the analogous formula
SWz = a SWz + b SWz
tT " P
holds if (J does not contain o.
Finally, for the form W3 we obtain similarly
S W3 = a S W3 + b S W3 + 2nin, n E 7L , (3)
tT " P
because W3 has simple poles at p and 0, and by analogy to the expansion (2)
W3 = i + h du, h E (!) P' where u is a local parameter at p.
(! (!
The vectors Wl' ~ Wl) and wz, ~ wz) are not proportiorral. For
if a combination of them with coefficients A and )1 would vanish, then we
would have the relation
z
for every cycle (J. This means that the function cp(x) = S(AWl + )1W z) (for
some fixed point q) is single-valued and meromorphicq on Can. By Theo-
rem 1 it must be a rational function on C. If )1 # 0, then it has only one
simple pole at 0, which is impossible, because C is not a rational curve.
And if )1 = 0, then it is regular everywhere, which is also impossible.
Utilizing the independence of these vectors we can find numbers A
and )1 such that
(J J =)'(J
W 3, W3) Wl'} Wl) + )1(1 w 2 , Iwz).
We set 11 = Oh = AWl - )1W z . The equation (3) shows that
SI1=2nin, nE7L,
x
for every cycle (J. Therefore the function cp(x) = exp S11 is single-valued
q
on Can. It is meromorphic on Ban and regular everywhere except possibly
at p. In a neighbourhood of this point (1)3 = (~ + h) du, hE (9 u' and 11 has

the same expansion, hence <p = u .11', where 11' is holomorphic and dif-
ferent from 0 at p. This shows that the divisor of <p on Ban consists of the
point p with coefficient 1.
3. Example of a Non-Algebraic Compact Manifold with the Maximal

Number of Independent Meromorphic Functions. The transcendence
degree of .A(X) (which is finite by Theorem 3) is the basic invariant by
means of which we might try to classify compact analytic manifolds.
Here we report, omitting all proofs, about what is known in this direction.
From this point of view the manifolds closest to algebraic varieties
are those for which the transcendence degree of .A(X) is equal to the
dimension of X. We begin by constructing an example ofa non-algebraic
manifold with this property.
We use a construction very similar to the one we have used in Ch. VI,
§ 2.3, to construct an example of a non-projective algebraic variety.
Here we apply the concept of a u-process along a smooth subvariety
to the case when the enveloping manifold is analytic. The reader can
easily verify that the definitions and simplest properties deduced in
Ch. VI, § 2.2, carryover word for word to this case.
We consider the projective space ]p3 and in it a curve C having a
double point Xo with distinct tangents, for example, the curvi! with the
equation (1) in Ch. I, § 1.1. There exists a neighbourhood U of Xo (in the
complex topology of ]P3) such that the analytic manifold Un C is
reducible and splits into two one-dimensional irreducible smooth sub-
manifolds C and C" intersecting transversally, namely, the two branches
of C at Xo.
We consider the u-process of the variety U, u 1 : U1 - U with centre
in the subvariety c.
The inverse image C 1 = ul1(C) is a smooth surface,
with fibres isomorphic to ]P 1. We set u - 1(xo) = L 1 • The inverse image
U- 1 (C") is reducible and consists of two one-dimensional components:
L1 and a smooth subvariety C~, which is mapped by u, isomorphically
onto C". Both subvarieties intersect transversally at the point x 1 = L1 nC~.
We now consider the u-process U2 : V - U1 ofthevariety U1 with centre in
C~. Again the inverse image u;1(L1) consists of two one-dimensional
components: u;1(L 1)=IuI 1, where I=u;1(x1)' and where I1 is
mapped isomorphically by U 2 onto L 1 • We set (;'=U2.U1: V-U. On
the other hand, we consider the u-process u: V _]p3 - Xo of the variety
]p3 _ Xo with centre in the subvariety C - Xo. Since u coincides in U - Xo
with the u-process in C - xo, the two varieties and mappings we have
constructed can be combined into a single one:
Evidently <C(JP3) C A(X), hence the transcendence degree of the field

A(X) is 3. We show that X is not an algebraic variety. To do this we
assume that it is algebraic, and for the curve situated on it we use the
concept of numerical equivalence introduced in connection with the
similar example in Ch. VI, § 2.3. We also use the fact that on an algebraic
variety an irreducible smooth curve is not equivalent to zero. For as we
have seen in Ch. VI, § 2.3, for this purpose it is sufficient to construct an
effective divisor intersecting our curve in a non-empty finite set of points.
Let E C X be our curve and U C X an affine open set (it is here that we
assume X to be algebraic) having a non-empty intersection with E.
In U we can find a divisor intersecting Un E in a finite and non-empty
set of points, for example, a hyperplane section F in the enveloping affine
space, passing through some point x E U nE but not through another
point x' E U nE. The closure F of the divisor F in the whole of X then
has all the properties we need.
Now it is sufficient to find on X an irreducible curve equivalent to
zero so as to obtain a contradiction to the fact that X is an algebraic
variety. For this purpose we use the result that under a a-process with
centre on a curve the inverse images of all points of this curve are
equivalent to each other. We take the points x E C - X o, x' E C' - X o,
x" E C" - X o, and let L = a- 1(x), L' = a- 1(x'), L" = a- 1(x"). Considering
L" as inverse image of a 11 (x") under a 2 we find that
L~L"~I. (1)
On the other hand, on U 1
but on []
Thus,
L~I+I1·
In conjunction with (1) this shows that I1 ~ o.

Observe that in all these arguments we could have considered instead
of equivalence of curves on X homology of the corresponding cycles, using
the results of Ch. VII, § 1.3.
The dimension 3 in our example is the smallest possible, because it
can be shown that a compact analytic manifold of dimension 2 on which
there exist two algebraically independent merom orphic functions is an
algebraic variety, hence even projective, as was indicated in Ch. VI, § 2.3.
Analytic manifolds X for which the transcendence degree of ...tt(X) is
equal to the dimension of X are very close to algebraic varieties. In this
case .,H(X) is isomorphic to the field <C(X') ofrational functions on some

algebraic variety X', dim X' = dim X, so that X is "bimeromorphically
isomorphic" to an algebraic variety. This fact can be made more precise
by proving for such varieties an analogue to Chow's lemma in Ch. VI,
§ 2.1. All this suggests that for such manifolds there exists a purely
algebraic description and that the analogous objects can be defined for an
arbitrary field. In fact, such a concept, a so-called "algebraic space", has
recently been introduced. The reader can get acquainted with them in the
papers [5] and [22].
4. Classification of Compact Analytic Surfaces. In our classification

we now go over to the type of manifolds for which the transcendence
degree of .,H(X) is dim X - 1. Owing to Riemann's existence theorem this
case is impossible for dim X = 1, and we can expect to meet it first for
dim X = 2, that is, for analytic surfaces. Examples of such surfaces are
known to us. They are the complex tori in Example 1 of § 1.4 and the
Hopf manifolds (Example 3 there). A general classification of them is
given by the following theorem of Kodaira:
A compact analytic surface X for which the field .,H(X) is of trans-
cendence degree 1 has a holomorphic mapping p : X -+ Y onto an algebraic
curve Y such that .,H(X) = p*<C(Y) and all its fibres p-1(y), except finitely
many, are elliptic curves.
A similar fact can be proved for manifolds of arbitrary dimension,
but in a weaker form:
If X is a compact n-dimensional analytic manifold and the trans-
cendence degree of .,H(X) is n - 1, then X is bimeromorphically isomorphic
to a manifold X' having a holomorphic mapping p: X' -+ Y onto an (n - 1)-
dimensional algebraic variety Y such that .,H(X) = .,H(X') = p*<C(Y),
and p-1(y) is an elliptic curvefor all points y in some set that is open in Y
in the spectral topology.
Other types of analytic manifolds have been investigated almost
exclusively in the case of analytic surfaces. The only remaining type for
them is .,H(X) = <C. Here we describe a classification ofthis type of surface
due to Kodaira.
First of all, we note that the concept of an exceptional subvariety
naturally carries over to analytic manifolds. It can be shown that every
analytic surface can be obtained by finitely many u-processes from a
surface not having exceptional curves. In this context we are going to
talk simply of surfaces without exceptional curves.
Kodaira has shown that for a compact surface X without exceptional
curves for which .,H(X) =<C the one-dimensional Betti number b 1 can
only assume the three values: 4,1, and O.
If hl = 4, then X is a complex torus. We already know an example of

a complex torus on which all meromorphic functions are constants
(Example 2 to § 1.4).
If hl = 0, then the canonical bundle of the surface is trivial (the
canonical bundle is defined by analogy with the case of algebraic
varieties and is a substitute for the canonical class when we cannot use
rational or merom orphic functions, see Exercises 3 and 4 to § 2). All
surfaces of this type are homeomorphic to each other and to algebraic
surfaces of type K 3 (see Ch. III, § 5.7). They are called analytic surfaces
of type K3.
The case hl = 1 has not yet been investigated completely. Examples of
such surfaces can be constructed by generalizing the construction of a
Hopf variety. Namely, compact surfaces having the form ([:2 - O)/G,
where G is a discretely and freely acting group of automorphisms of
([:2 -0 are called generalized Hopfmanifolds. For example, for G we can
take the cyclic group generated by the automorphism (z l' Z 2)-+ (0( 1Z 1, 0(2 z 2),
10(11 < 1, 10(21 < Lit can be shown that if there do not exist integers nl andn2'
not both 0, for which O(~j = 0(22 , then on such a surface all meromorphic
functions are constant.*
Thus, with respect to the invariant quantity 1, the transcendence
degree of At(X), compact analytic surfaces can be classified as follows:
1 = 2 - algebraic surfaces;
1 = 1- surfaces with a sheaf of elliptic curves;
1 = 0 - the surfaces are tori or are of type K3 or have hl = 1 (the
classification of all such surfaces is an interesting problem).
In this classification a striking resemblance with the classification
of algebraic surfaces, as explained in Ch. III, § 5.7, hits the eye. An
understanding of this analogy is probably possible only in connection
with a generalization of the two theories to manifolds of arbitrary
dimension. This is one of the most interesting tasks of the theory of
algebraic varieties and analytic manifolds.
Exercises
1. Let A = ([;110 be a two-dimensional complex torus, g the automorphism gx = - x,
G = {1, g}. Show that the ringed space X = AIG (see Exercise 1 to Ch. V, § 3) is a complex
space having 16 singular points Zj" •. ,Z16 corresponding to the points xeA for which
2x=O.
* Recently Inue has constructed a series of such surfaces, different from~generalized
Hopf manifolds. Their second Betti number is O. Bogomolov has proved that all surfaces
with b1 = 1, b1 = 0 are covered by Inue's construction. Finally, Hirzebruch constructed
surfaces witlt b1 = 1, b1 > O. At present there are no conjectures about their structure.
See E. Bombieri, D. Husemaller: "Classification and embeddings of surfaces". Proceed-
ings of Symposia in Pure Math., Vol. XXIX and F. Bogomolov, Izv. Acad. Nauk USSR,
Ser. Matlt., Vol. 40, 1976, N2. (Footnote to corrected printing, 1977).
2. In the notation of Exercise 1, show that every singular point ZI E X has a neigh-
bourhood isomorphic to a neighbourhood of the singular point of a quadric cone (see
Exercise 8 to § 2).
3. In the notation of Exercises 1 and 2, show that there exists a complex manifold X
and a holomorphic mapping ({I:X ->X such that on X there are 16 pairwise disjoint curves
C 1 , ... ,C 16 each of which is isomorphic to lP;o, ({I(C1)=z;, and ({I:X-UCi .... X-UZi
is an isomorphism. (Hint: Use Exercise 10 to Ch. II, § 4.)
4. In the notation of the preceding exercises, show that the differential form dz 1 /\ dz 2 ,
where z 1 and Z2 are coordinates in ([:2, determine a differential form on A that is holo-
morphic and vanishes nowhere (see Exercises 3 and 4 to § 2). Show that it also determines
a holomorphic form on X vanishing nowhere. Deduce that the canonical bundle on X is
trivial.
5. In the notation of the preceding exercises, show that if all the meromorphic functions
on the torus A are constant, then the same is true for X. Show that X is not Isomorphic to
a complex torus (for example, verify that on X there are no one-dimensional holomorphic
differential forms). Thus, X is an example of a non-algebraic surface of type K 3.
6. Show that for any smooth projective variety X of dimension n ~ 3 there exists a
non-algebraic complex compact n-dimensional manifold X such that ..Jt(X') = ([:(X).
Chapter IX. Uniformization
§ 1. The Universal Covering

1. The Universal Covering of a Complex Manifold. In the preceding
sections we have made use of the concept of factor space to construct many
important examples of analytic manifolds. Now we show how this
concept leads to a general method of studying such manifolds.
We begin by recalling some simple topological facts (see, for example,
[25J, §§ 49-50). Let X be an arcwise connected, locally connected, and
locally simply-connected space. Later X will be a connected manifold,
and all these conditions hold. The universal covering space X of X has a
projection
p:X-+X,
which turns X into an unramified covering. The homeomorphisms g
of X into itself for which p. g = p form a group G, which is isomorphic
to the fundamental group 1t 1 (X) of X. The group G acts on X discretely
and freely, and
( 1)
Now we assume that X is an analytic manifold, and we denote by (9x
its structure sheaf. The manifold X (and more generally, any unramified
covering) can also be turned into an analytic manifold, in fact, so that the
projection p is a holomorphic mapping. To see this we consider the
presheaf & on X defined by the condition
&(U) = (9x(P(U))
for U open in X [then p(U) is also open in X, because p determines an
unramified covering]. The sheaf associated with & is denoted by (9 x.
Every point xE X has a neighbourhood U that is mapped by means of p
homeomorphically onto p(U). Therefore the sheaf (9 xis given uniquely by
its restrictions to these sets U. It is easy to see that for them (9x is obtained
simply by transferring the sheaf (9x by means of the homeomorphism p.
From what we have said it follows that the pair (X, (9x) determines an
analytic manifold. For if p: U -+ p(U) is a homeomorphism, then the
projection p determines an isomorphism of the ringed spaces (U, @ilu)
§ 1. The Universal Covering 381
and (P(U), (9xlp(u))' Therefore, if p(U) is isomorphic to a domain in <en,

then the same is true for U. It is also evident that p is a holomorphic
mapping. Furthermore, since the complex structure on X is defined by
means of the projection p, and since the homeomorphisms g E G do not
change under this projection, they are automorphisms of the analytic
manifold X. Hence it follows that (1) is an isomorphism of analytic
manifolds.
Suppose that two manifolds X and X' have a common universal
covering X. Then
X = X/G, X' = X/G'
and there are two unramified coverings p : X~ X and p' : X~ X'. Let us
find out when X and X' are isomorphic. Here we use an elementary
topological fact (which justifies the term "universal" covering): ifp: X~ X
is a universal covering and q : Xl ~ X any connected unramified covering,
then there exists a continuous mapping CP:X ~X I such that qcp = p.
Let f: X' ~ X be an isomorphism. Then q = f p' determines an unramified
covering q : X ~ X. By applyi~g the results stated above we construct a
continuous mapping cp : X ~ X such that the diagram
- tp -
X~X
p'l
X/~X
J 1p
commutes. Hence it follows that cp is a holomorphic mapping. For it
follows from the commutativity of the diagram that pcp is holomorphic,
that is, for a function u E (9 x, x the function (pcp)* u = cp* (P* u) is holo-
morphic at the points x ElPCP)-I(X)=cp-l(p-I(X)). But all functions
holomorphic in a neighbourhood of x E p-1(X) have the form p*(u)
locally, and this implies that cp is holomorphic. By changing the roles
of X' and X we see that cp is an automorphism of the analytic manifold X.
We recall that the groups G and G' consist of the automorphisms of
X for which
py=p for yE G
and
p' y' = p' for y' E G' .
Multiplying the second equality by f and using the commutativity of the

diagram (2), we see that G = cpG' cp - 1.
So we have proved the following result:
Theorem 1. Any connected analytic manifold can be represented in the
form (1), where X is a simply-connected analytic manifold and G is a
discretely and freely acting group of automorphisms of it. In all such
382 Chapter IX. Uniformization
representations of one and the same manifold X the groups G are con-
jugate in the group of all automorphisms of X.
2. Universal Coverings of Algebraic Curves. Theorem 1 enables us to

reduce the study of arbitrary analytic manifolds to that of simply-
connected manifolds and their groups of automorphisms. Of course, the
problem is only shifted in this way-everything depends on how much
we know about simply-connected analytic manifolds. In general this is
very little; more details on this will be in § 4. An exception are one-
dimensional manifolds, and from now on we confine our attention
mainly to them.
Classification of connected simply-connected one-dimensional ana-
lytic manifolds is very simple. There are altogether three:
1) the projective line WIn;
2) the affine line A!n = ([:1 ;
3) the interior of the unit disc D, defined in ([:1 by the condition Izl < l.
[The manifolds 1) and 2) are known in the theory of analytic functions
as the Riemann sphere and the finite plane.] This theorem can be proved
by the same methods as Riemann's existence theorem. A proof can be
found, for example, in the book [31], Ch. 9, § 1.
It is easy to verify that the three manifolds in the theorem are not
isomorphic. The first of them is not isomorphic to the second or the
third, because it is compact and they are not. The third is not isomorphic
to the second because on it there exist bounded holomorphic non-
constant functions, whereas by Liouville's theorem on the second there
are none.
Thus, all connected one-dimensional analytic manifolds fall into three
classes depending on which of the three types their universal coverings
belong to. The types 1), 2), and 3) are called elliptic, parabolic, and
hyperbolic manifolds; this terminology also applies to non-compact one-
dimensional analytic manifolds.
In order to investigate manifolds of these three types we have to know
the discretely and freely acting groups of automorphisms of their universal
coverings. The answer follows easily from simple facts of the theory of
analytic functions of a single complex variable.
Proposition. Every automorphism of the manifold WIn has a fixed point.
A discretely and freely acting group G of automorphisms of ([:1 for which
([:1/G is compact consists of the translations z -> z + a, where a ranges
over the vectors of a two-dimensional lattice on ([:1. All the automorphisms
of the unit disk have the form
z-Ct.
z->o--_-, 101 = 1, 1Ct.1 < 1. (1)
1-Ct.z
According to Theorem 2 of Ch. VIII, § 3, every automorphism of IP~n

is of the form gan' where g is an automorphism of the algebraic variety IP 1,
hence is a fractional linear transformation. Since every fractional linear
transformation has a fIxed point, the fIrst assertion of the proposition
follows from this.
An automorphism of ([:1 is given by an entire function fez). If this
function had an essential singularity at infInity, then it would assume in
any neighbourhood values arbitrarily close to every given number
(Weierstrass' theorem). This contradicts the fact that / determines an
automorphism. For if f(a) = b, then / assume all values sufficiently close
to b in some neighbourhood of a and cannot assume them in a neighbour-
hood of 00. Thus, / is a polynomial. If its degree is n, then it takes every
value n times. Therefore / determines an automorphism only for n = 1.
In other words, every automorphism of ([:1 is of the form
fez) = az + b, a =1= o. (2)
The automorphisms occuring in a freely acting group G do not have

fIxed points. Hence for them a = 1 in (2). So we see that G must consist
of the translations fez) = z + b. If we make use of the group structure in
([:1, we can restate our result by saying that G is a subgroup of ([:1, and X
the factor group ([:l/G.
In Ch. VIII, § 1.4, we have already used the simple theorem that
determines all discrete subgroups G C ([:1 with compact factor group. In
our case it shows that G must coincide with a two-dimensional lattice
7lW1 + 7lW2' where W1' W2 E ([:1 are independent over lR.
Finally, let D be the interior of the unit disc. Substitution shows
that the transformations (1) form a group and that this group acts
transitively on D. Therefore, when we multiply any automorphism by
some automorphism (1) we can obtain an automorphism y leaving the
point 0 fIxed. It is therefore sufficient to show that these automorphisms
are of the form (1). If y(O) =0, then by Schwarz's lemma (Ch. VIII,
§ 2.3) in D
Iy (z)/z I ~ 1 ,
and since y(z) and z are symmetrical, we also have Iz/y(z)l..; 1, hence
Iy(z)/zl = 1. From this it follows that the functio'n y(z)/z is constant:
y(z) = Oz, 101 = 1 .

This proves the proposition.
Thus, the classifIcation of manifolds of elliptic type is trivial: they
are all isomorphic. For in the representation (1) of § 1.1 for them X = IPin>
also G=e and X =X =1P~n.
Compact manifolds of parabolic and hyperbolic types are worked out

in the next two subsections. We show that for every discretely and freely
acting group G for which X/G is compact this manifold is a projective
algebraic curve, and we construct an explicit projective embedding of
these manifolds. In this way we give a proof of Riemann's existence
theorem, starting out from the classification of simply-connected one-
dimensional manifolds. Furthermore, we show that compact manifolds
of parabolic type coincide with algebraic curves of genus 1 (elliptic curves),
and manifolds of hyperbolic type with curves of genus greater than 1.*
3. Projective Embeddings of Factor Spaces. In several special cases

we have had to investigate the following general situation. Let X be a
one-dimensional analytic manifold, and G a freely and discretely acting
group of automorphisms of it. We assume that the factor space X = X/G
is compact; how can we construct an immersion of it in a projective
space IP"?
We shall give such an immersion by n + 1 functions fo, ... , In that
are holomorphic on the whole of X. We assume that they do not vanish
simultaneously at any point x E X. Then
(1)
is a holomorphic mapping.
For f to be a mapping of X into IP" we could require invariance of the
functions fi under all g E G. But then these functions would be holo-
morphic on X, hence constant by Theorem 2 of Ch. VIII, § 2. However,
this condition can be weakened by requiring only that for every g E G
there should exist a function ({)g on X such that
g* fi = fi({)g, i = 0, ... , n. (2)
It then follows thatf·g = ffor all g E G, hence f can be decomposed:

f = J-n, where n is the projection X ~ X and f some holomorphic
mapping X ~IP". We say that f determines the mapping f: X ~IP".
From the fact that the functions fi are holomorphic and do not vanish
simultaneously on X it follows that the functions ({)g are holomorphic and
do not vanish anywhere on X.
Let us clarify when this system of functions determines an isomorphic
embedding f : X ~ IP".
Proposition. Let X be a one-dimensional analytic manifold, G a dis-
cretely and freely acting group of automorphisms of it, fo,···, In functions
* The terminology is clearly very bad: elliptic curves belong to the parabolic type, and the
projective line to the elliptic type! But it has been universally accepted for such a long time
that we dare not change it.
that are holomorphic on .:f

and satisfy (2), where qJg is a holomorphic
function without zeros on X.
We assume the following conditions to be satisfied:
A) r (fo(x'), ... , fn(X')) = 2

g fo(x"), ... , fn(x")
for all points x', x" E X, provided that x" 1= gx' for all g E G, and
B) rg(fo(X), ... , f,,(X)) =2

f~(x), ... , f~(x)
for all x E X [f'(x) denotes the derivative of f as a function of the local

parameter at x, and condition B) does not depend on the choice of this
parameter]. Then the mapping (1) determines an isomorphic embedding
of the manifold X =XjG in IPn.
The proof reduces to a simple verification. Condition A) guarantees
that all the functions fi do not vanish simultaneously at any point
x E X, so that (1) in fact gives a point of the projective space. Condi-
tion (2) shows that f determines a mapping
f: X -+1Pn,
which is holomorphic by the preceding remark. Condition A) guarantees

that it is one-to-one.
Suppose that fo(x o) 1= 0 for some Xo E X. The corresponding point
Xo E X has a neighbourhood U in which f is given by equations
Yi = gi(X) = fi(x)j fo(x), i = 1, ... , n ,

where Yl, ... , Yn are coordinates in the affine space JAn into which U is
mapped. From condition B) it follows that for some i> 0 we have
g;(xo) 1= O. We assume that i = 1, that is, g'l (x o) 1= O. By virtue ofthis we may
express the local parameter z at Xo as an analytic function of Y 1 = g 1(z):
So we see that f(X) in a neighbourhood of f(x o) is given by analytic

equations
Yi - gi(h(Yl)) = 0, i = 2, ... , n,
where the functions U 1 = Yl, ui=gi(h(Yl)) - Yi, i= 2, ... , n, form a system

of local coordinates in a neighbourhood of f(x o) in 1Pn. This shows
that f(X) is an analytic submanifold in 1Pn.
Finally, the mapping inverse to I is given in a neighbourhood of

I(zo) by the function z = h(Y1) (we recall that z can be regarded as a local
coordinate on X). Therefore I is an isomorphic embedding. This
Exercises
I. Show that the universal covering of an n-dimensional Abelian manifold over ([ is
isomorphic to ([n.
2. Show that birationally isomorphic smooth projective surfaces have isomorphic
fundamental groups.
3. Let X be a compact analytic manifold. Show that there exists only finitely many
non-isomorphic manifolds Y having a holomorphic mapping f : Y ... X that tum Y into
an unramified covering of X of given degree m.
4. Determine the fundamental group and the universal covering X for the manifold
X = 1P 1«[) - (0) - (00), and find a representation X = X/G, where G is a discrete group of
automorphisms of X.
5. The same as in Exercise 4 for X = D - (0), where D = {z, Izl < 1}.
6. Show that the universal covering of the manifold X =IP1(<C)- ce - P-)I. where
ce, P,)I E IP 1«[) are three distinct points, is isomorphic to D. Use the classification of simply-
connected one-dimensional manifolds given at the beginning of § 1.2.
7. Deduce from the result of Exercise 6 Picard's theorem: if an entire function f
does not assume two values ce and p, ce # p, then it is a constant. Hint. Interpret f as a
mapping ([1 ... IP 1«[) - ce - p - (00).
§ 2. Curves of Parabolic Type

1. 8-Functions. From the proposition in § 1.3 it follows that any compact
manifold of parabolic type is a one-dimensional torus, that is, has the form
([1 jQ, where Q is a two-dimensional lattice. According to Theorem 1 of
§ 1, two lattices Q and Q' lead to isomorphic factor spaces if and only
if the groups of translations corresponding to them are conjugate under
some automorphism I of ([1. Evidently this is equivalent to the fact that
Q' = IQ. Since J must be expressible in the form J(z) = az + b, it follows
that the lattices Q' and Q are similar.
Our aim is to show that everyone-dimensional torus X is of the form
Y..n, where Yis a projective curve. For this purpose we use the method that
was described at the end of the preceding section. First of all, since we
can replace the lattice by a similar one without changing the torus ([1/D,
we may assume that it has the basis 1, 'I: with 1m 'I: > O. We try to construct
an embedding of ([1 jQ in IPn by means of functions 10"'" in satisfying
the following special form of the relations (2) of § 1.3:
fi(z + 1) = Ii (z),
fi(z + '1:) = e- 2"ilz fi(z) (1)
i=O, ... ,n,
where I is a positive integer. Formally speaking, this choice does not

require a justification, if only we can show that for some I we can find
linearly independent functions satisfying the relations (1) and the condi-
tions of the proposition in § 1.3. However, it can be shown that~ in fact,
the functions giving any mapping of X in ]pn reduce to this form. The fact
of the matter is that we do not change the mapping when we multiply all
the functions /;(z) by eu(z), where u(z) is an entire function. It is not hard
to show that by making use of this we can always put the relations (2) of
§ 1.3 into the special form (1).
Defmition. Entire functions satisfying the condition (1) are called
{}-functions of weight I.
Clearly all (}-functions of equal weight form a linear space, which we
denote by 5e/.
Theorem 1. The dimension of the space 5e/ is I.
Proof The first of the conditions (1) shows that f(z) = cp(t), where cp is
a function holomorphic in ([:1 - 0 and t = e2niz . For cp(t) = f( 2~i log t)
is a single-valued analytic function in ([:1 - O.
Let
L
(J()
cp(t) = cmt m
m=-oo
be the expansion of this function in a Laurent series. Setting e2ni< = A

we can rewrite the second condition (1) in the form
L cmt mAm = L cmtm-/ = L cm +/ tm ,

meZ meZ meZ
or
(2)
We set
m=l·r+a, O<,.a<l. (3)
From the relations (2) we find that

c =C Ara+/ r(r; I)
m a .
Thus, the function cp is given uniquely by the numbers co, ... , c/- 1 , from
which it follows that dim 5e/ <,. I.
To complete the proof of the theorem it is sufficient to show the
convergence of the series corresponding to arbitrary sequences satisfying
the relations (2). We may restrict ourselves to the single arithmetic

progression (3). Then we obtain the series
r(r-l)
cat" I urJJ.-z-
rE71.
where u = tl Aa , JJ. = AI. From the condition 1m r > 0, I> 0, it follows

that IJJ.I < 1. Therefore the convergence of the series
r(r-l)
IlulrlJJ.l-z-
is obvious, and the theorem is proved.
Note. From the theorem it follows that to within a factor a O-function
of weight 1 is unique. If in (2) we set Co = 1, this function is uniquely
determined. It is denoted by O(z).
2. Projective Embedding. Now we can prove the main result of this

section.
Theorem 2. The O-functions of weight I > 3 determine an isomorphic
embedding of the manifold X =<C 1 jQ.
We give the proof for 1= 3; the general case is completely similar.
We make use of the following obvious remark. If f(z) is a O-function of
weight I, and a 1 , •.. , am complex numbers with a 1 + ... + am = 0, then
n f(z + ai)
m
g(z) =
i= 1
is a O-function of weight I m. In particular, for arbitrary a and b

f(z) = O(z + a) O(z + b) O(z - a - b)
is a O-function of weight 3.
We have to verify that the conditions A) and B) of the proposition in
§ 1.3 hold for three linearly independent O-functions of weight 3. If A) does
not hold for three basis functions of 2 3 , then there exist a and {3, not
both zero, and z' and z" such that z' - z" ¢ Q, and af(z') = {3f(z") for any
function f E 2 3 , In particular,
aO(z' + u) O(z' + a) O(z' - u - a) = {30(z" + u) O(z" + a) O(z" - u - a)
for arbitrary u and a. We set z' + u = z, z" - z' = Co and we regard z as

variable and the remaining quantities as fixed. So we see that
aO(z) O(z' + a) 0(2z' - a - z) = {30(z + () O(z" + a) O(z' + z" - a - z) .
We choose a so that the functions O(z) and lJ(z' + z" - a - z) do not have
common zeros. Then the functions O(z + () and O(2z' - a - z) have the
same property. Therefore O(z)jO(z + () has no zeros nor poles, from which
it follows that lJ(z + () = eg(z) O(z), where g is an entire function. From
the definition of O(z) it follows that
g(z + 1) = g(z) + 2lni, (1)
g(z + -r) = g(z) - 2ni( + 2nil', I, l' E Z. (2)
Thus, the function g'(z) has the periods 1 and -r, therefore, it is bounded
on ([;1, and since it is entire, it is a constant. So we see that g(z) = (XZ + /3,
and from (1) it follows that (X = 21ni, and from (2)
2lni-r = - 2ni( + 2nil' ,( = I' - 1-r E Q.
This shows that condition A) in the proposition of § 1.3 holds. Condition
B) is verified similarly. Namely, if it does not hold, then there exists a
Zo E ([;1 such that (1'1 !)(zo) =0 for all ! E 2 3 . In particular, we can set
! = O(z + u) lJ(z + a) O(z - u - a). We find that
lJ'(zo + u) O'(zo + a) O'(zo - u - a) 0
--::-::""'::'----,'- + + = . (3)
lJ(zo + u) O(zo + a) lJ(zo - u - a)
Again we regard u as variable and we choose a so that the functions
lJ(zo + u) and lJ(zo - u - a) do not have common zeros. Here (3) is possible
only if all three terms on the left-hand side are entire functions of u. This,
in its tum, is possible only if lJ(z) has no zeros, that is, O(z) = eg(z), where
g(z) is an entire function. This representation at once leads to a contradic-
tion to the definition of lJ. For it follows from this definition that
g(z + 1) = g(z) + 2mni, (4)
g(z +-r)=g(z)- 2lniz+ 21' ni. (5)
Hence we find, as above, that g"(z) is constant. Therefore g(z) = (XZ2 + /3z + ,)"
and from (4) we see that (X =0 and from (5) that 1= 0 against the condition
1> O. This proves the theorem.
Note. In a similar but more complicated manner it can be shown that
a many-dimensional torus is projective if its period matrix satisfies the
Frobenius relations of which we have talked in Ch. VIII, § 1.4.
3. Elliptic Functions, Elliptic Curves, and Elliptic Integrals. Having

constructed the mapping
!: X-+IP" ,
we are interested in studying it in more detail. We have seen that f(X)

is a smooth algebraic curve. The addition of points on the torus deter-
mines a group structure on· Y. Here the addition mapping
/1:YxY-Y
determines a holomorphic mapping of the corresponding complex
manifolds. According to Theorem 2 in Ch. VIII, § 3, it follows that /1 is
a morphism. Thus, Y is a one-dimensional Abelian variety. In Ch. III
we have seen that in this case the canonical class of Y is 0, hence the
genus is 1. So we have shown that compact manifolds of parabolic
type are smooth projective curves of genus 1 (elliptic curves) and only they.
Note that it makes sense to talk of the zeros of O-functions on X:
although the value of O(z) changes when z is replaced by z + a, a E Q, if
O(z) = 0, then also O(z + a) = O. The usual definition allows us to talk
of the divisor of a O-function on X.
O-functions of weight 3 lead to an isomorphic embedding of X in 1P2.
In this case Yis a smooth plane curve of genus 1. From the formula for
the genus of a plane curve it follows that the degree of Y is 3. In particular,
every O-function of weight 3 determines on X a divisor, and the mapping
f: X _1P 2 assigns to it the divisor of the section of Y = f(X) with a
line in ]p2, which is of degree 3. Applying this remark to the function 03 ,
where 0 is of weight 1, we see that the divisor of 0 on X consists of a single
point with multiplicity 1. If this point is x o, then O(z - a + xo) has the
divisor a. This sheds new light on the role of the O-functions. If we make
use of the O-functions (which, of course, are neither meromorphic nor
functions on X, in general), then every divisor becomes principal.
The embedding f we have constructed determines an isomorphism
of the fields ([:(Y) and '#(X). On the other hand, '#(X) can be described
as the field of meromorphic functions on ([:1 having the periods 1 and T.
Such functions are called elliptic. Thus, the field ([:( Y) is isomorphic to the
field of elliptic functions. In particular, if
F(x,y)=O
is the equation of an affine model of the curve Y, then there exists· a

parametrization
x = cp(z), y = 1p(z)
by elliptic functions. This parametrization is called a uniformization of

Y. In this way a connection is established between elliptic functions and
elliptic curves: the former uniformize the latter.
Let Y be an elliptic curve. How can we find a lattice Q corresponding
to it for which Y = ([:1 /Q? Let w be a regular. differential form on Y. Since
the genus of Y is 1, it is uniquely determined to within a constant factor.

If I: ([;1 ~ Y is the holomorphic mapping we wish to find, then I*w is a
hoi om orphic differential form on ([;1, which must be invariant under
translations by vectors of Q. This means that 1*(1) = u(z) dz, where u(z)
is an entire function that is invariant under translations in Q. Hence it
follows that u is constant and, using the fact that w is arbitrary, we may
assume that
I*w=dz.
Let Zo E Q, that is, I(zo) = 1(0), and let s be a path joining 0 to Zo
in ([;1. Then
Zo
Zo = Jdz = JI*w = J w. (1)
o J(s)
The path I(s) is closed on Y(([;), so that it determines an element of the

group H1 (Y(([;), Z). Clearly all the elements of this group are obtained in
this manner. Formula (1) shows that the lattice Q coincides with the set
of complex numbers
In particular, a basis of it consists of the numbers

Jw, J
(1) ,
where 0"1 and O"z form a basis of H1(y(([;), Z). For example, if the curve Y
is given by an equation V Z = u 3 + Au + B, we may set (1) = dujv, and the
basis of Q consists of the numbers
du J du ,
L (u 3 +Au+B)t' G2 (u 3 + Au + B)2
The integrals S ware called elliptic integrals, and S w, where

G G
0" E H1(Y(([;), Z), are called their periods. Thus, the lattice Q, which
determines the torus ([;ljQ isomorphic to the elliptic curve Y, consists of
periods of the elliptic integral connected with this curve.
In conclusion we mention that the uniformization of elliptic curves
puts us in a position to understand from a new point of view the funda-
mental fact of which we talked in Ch. III, § 5.6: not all curves of genus 1 are
isomorphic to each other. We can even form some idea of the structure of
the set of equivalence classes of elliptic curves.
To do this we represent any elliptic curve in the form ([;1 jQ, and
replacing, if necessary, the lattice Q by a similar one we choose in it a
basis 1, r, 1m r > O. It is easy to see that two such bases 1, rand 1, r'
determine similar lattices if and only if there exist integers a, b, c, d

such that
a-r+b
-r'= ad-bc=1. (2)
c-r+d'
The set of all transformations of the form (2) is a group G, the so-called
modular group. We denote by H the upper half-plane 1m -r > O. Since
elliptic curves are isomorphic if and only if the lattices corresponding to
them are similar, the set of isomorphism classes of elliptic curves is in
one to one correspondence with the points of the factor space HjG.
It can be shown that the group G acts on H discretely, but not freely
(it has fixed points). Nevertheless, the factor space HjG is a one-
dimensional complex manifold. Moreover, it is isomorphic to ([:1. The
functionj: HjG-+([:l that realizes this isomorphism establishes a one-to-
one correspondence between classes of isomorphic elliptic curves and
complex numbers. An algebraic description of it can be extracted from
Exercises 12 and 13 to Ch. III, § 5.
Exercises
1. Show that if an elliptic curve X is determined by an equation with real coefficients,
then it is isomorphic to ([:l/Q, where Q = 71. + i71. or Q = 71. + 1; i 71., and in the first case
X(IR) consists of a single oval, in the second of two.
2. Show that if a real elliptic curve X consists of a single oval, then it is not homologous
to 0 on X«[:).
3. Show that if a real elliptic curve X consists of two ovals T1 and T2 , then for a suitable
orientation T1 and T2 are homologous in X(<C).
4. Show that all f)·functions of weight 0, 1, ... with given periods 1, T form a ring and
that this ring can be generated by the f)·functions of weight .;; 3.
5. Let f be an elliptic function with the period lattice Q. Show that the number of
zeros of f that are inequivalent modulo Q is equal to the number of i·ts poles that are
inequivalent modulo Q.
6. In the notation of Exercise 5, let 1X 1, ••• , IXm and P1' ... ' Pm' respectively, be inequivalent
zeros and poles of f Show that 1X1 + ... + IXm - P1 - ... - Pm E Q. Show that any numbers
with this property are the collection of zeros and poles of some elliptic function.
7. Let X = ([:1 IQ, X' = ([:110: be two elliptic curves. Show that the group Hom (X, X')
of homomorphisms (of algebraic groups) of X into X is isomorphic to the group of those
complex numbers IX E ([: for which IXQ C Q'.
8. Show that one can choose forms ro and ro', regular on X and X', such that the
number IX E ([: corresponding by virtue of Exercise 7 to a homomorphism f E Hom (X, X)
is defined by the condition f*ro' = IXro.
9. Show that for an elliptic curve X defined over ([: the ring End X = Hom (X, X) is
isomorphic either to 71. or to 71. + 71. "1, where "1 satisfies an equation "12 + ay + b = 0, a, b E 71.,
without real roots. In the second case X = ([:1 IQ, where the lattice Q is similar to an ideal
of the ring of numbers 71. + 71.)'.
§ 3. Curves of Hyperbolic Type 393
§ 3. Curves of Hyperbolic Type
1. Poincare Series. Let us consider the interior of the unit disc D

and a group G consisting of automorphisms of D, of which we assume that
it acts on D discretely and freely and that the factor space X = DIG is
compact. We construct an embedding of X into a projective space.
Just as in the preceding section, the construction of this embedding
relies on a study of functions f holomorphic in D and satisfying the
condition
g*(f) = f·qJg, g E G, (1)
where qJg is a holomorphic function that vanishes nowhere on D. From (1)

it follows at once that .
(2)
For an automorphism g we set
The rule for the differentiation of a compound function shows that equa-
tion (2) holds for qJg = jg' hence for qJg = j~ with any positive integer 1.
It can be shown that, in a sense, all solutions of equations (2) can be
reduced to qJg = j~. We shall consider only this case.
Defmition. A function f holomorphic in D and satisfying the relation
g*(f) = f·j~, g E G,
is called an automorphic form of weight 1 relative to the group G.

Our next aim is the construction of automorphic forms. For this
purpose we take an arbitrary function h holomorphic and bounded in D
and consider the series
I g*(h)j~. (3)
gEG
A series of this kind is called a Poincare series. If it defines an analytic

function, then a formal verification shows that this function is an auto-
morphic form. Thus, it remains to prove the following result:
Proposition. A Poincare series converges for I;;" 2 absolutely and
uniformly on any compact set KeD.
We use the following simple property of analytic functions.
Lemma. If a function f(z) is analytic in the disc Izl ~ r, then

1
If(OW ~ -2 S If(zW dx 1\ dy,
nr Izl ~r
z=x+iy.
Proof For any g, 0 < g < r,
Multiplying this equality by gdg and integrating with respect to g from

oto r we obtain
Therefore,
If(O)12=~1
nr
s f(Z?dXl\dYI~~
Izl;;iir
s If(zWdxl\dy.
nr Izl;;iir .
Proof of the proposition. Since h is bounded on D, it is sufficient for

us to show the convergence of the series I ~gll or even the series
gEG
(4)
The proof uses the fact that ~g12 is the Jacobian of g. We denote by
s(U) the area of a domain U determined by the Euclidean metric of the
plane ([:1 in which D is contained. Then
s(g(U)) = S dx 1\ dy = S UizW dx 1\ dy. (5)

gU U
The convergence of the series (4) follows at once from this remark.
For let U be a circle with centre at Zo and sufficiently small radius r so
that g(U)n U = 0 for g E G, g #- e. According to Lemma 1 and the remark
made above
".
L... ~izo)1
2 1"
.;;; -2 L.,
S.~iz)1 2 dx 1\ dy = -2
1 " ( ) s(D) 1
L., S g(U) ~ -2- = -2 '
gEG nr gEG U nr gEG nr r
(6)
which proves the convergence.
To prove uniform convergence we observe that gK1 and K2 for any
two compact sets K1 and K2 intersect only for finitely many elements
g E G. For according to the definition of a discretely and freely acting

group (Ch. VIII, § 1.2) any two points x and y have neighbourhoods U
and V such that gUn V = 0 for all g E G except possibly one. We take an
arbitrary point XE K l' and for any point y E K 2 we choose neighbourhoods
Uy of x and Yy of y such that gUy n Yy = 0 for all g E G except possibly one.
From the compactness of K2 it follows that there exists a neighbourhood
U of x such that gU nK2 = 0 for all g E G except finitely many. Our
assertion now follows at once from the compactness of K l'
Now we take a sufficiently small r > 0 so that the circle of radius r
with centre at an arbitrary point of K is contained in a compact set
K' C D. For every 8> 0 let C C D be a circle, sufficiently close to D, such
that K' C C and seD - C) < 8. We denote by q the number of those
elements g E G for which gK' nK' =I 0. For all g E G except finitely
many gK' C D - C. Denoting by 1:' the sum extended over these g we
find that as in the derivation of(5):
",,'. 2 1"", q q
L., 11g(z)1 .;;;; s(g(U)) < - 2 seD - C) < - 2 '
- 2 L.,
nr nr nr
from which the uniform convergence of the series (4) follows.
Note. We denote by M the multiplicative group of functions holo-
morphic on D and vanishing nowhere. It is a module over G with respect
to the action f --'> g*(f). Condition (2) agrees with the definition of a one-
dimensional cocycle. In the construction of a modular form by means of a
Poincare series we can recognize the idea of a proof of the so-called
Hilbert's Theorem 90 in homological algebra.
2. Projective Embedding. Now we can proceed to the main result.

Theorem. Let G be a group of automorphisms acting discretely and
fi-eely on D such that the factor space X = DIG is compact. Then there
exist finitely many automorphic forms of one and the same weight which
determine an isomorphic embedding of X in IP".
Of course, the matter concerns a verification of conditions A) and B)
of the proposition in § 1.3 for these forms. To begin with we aim at
satisfying them locally.
Lemma. Let z', z" E D be arbitrary points such that z" =I gz' for all
g E G. Then there exist automorphic forms fo and fl satisfying condition
A) of the proposition in § 1.3 for these points. For every point Zo E D
there exist automorphic forms fo and fl satisfying condition B) of the
same proposition at this point. In both cases we may assume that
We are looking for forms satisfying condition A) for the points Zl and
zit, in the form of Poincare series
(1)
From the convergence of the Poincare series it follows that lig(Z/)\ < 1
and lig(zlt)\ < 1 for all g E G except finitely many. Let g = e, g1'···' gN
be these excluded elements. Then
lig(zW ~o for z = Zl, zit, I~ 00 . (2)
We take functions hi> i = 0, 1, such that they satisfy the conditions
hi(gm(Z/)) =0, m= 1, ... ,N

hi(gm(zlt)) = 0,
hO(Z/) h1 (Z/) ho(zlt) h1 (zit) # 0,
hO(Z/) h1 (zit) - ho(zlt) h1 (Z/) # o.
Such functions can be found, for example, among the polynomials.
Then
J;(Z/) = hi(Z/) + hi(g(Z/))jiz/)' = hi(Z/) + ul')(z/),
where ul')(Z/) ~ 0 as 1~ 00, by virtue of (2). A similar relation holds

for zit. Hence it follows that
Io(Z/) 11 (Z/) Io(zlt) 11 (zit) # 0,

Io(z/) 11 (zit) - Io(zll) 11 (Z/) # 0
for sufficiently large I.

Now we construct functions satisfying condition B) of the proposition
in § 1.3. Again we are looking for them in the form (1). Let e, gto· .. , gN
be the elements g E G for which liizo) \ ~ 1. We take hi> i = 0, 1, so that
hi(gm(zO)) = hi(gm(zo)) = 0, i = 0,1, m = 1, ... , N ,

ho(zo) h1 (zo) # 0,
ho(zo) h'1 (zo) - h1 (zo) h~(zo) # o.
As before,
!i(zo) = h;(zo) + u11) (zo),

!i'(zo) = hi(zo) + V!l) (Zo), 'i=O, 1,
for sufficiently large t. This proves the lemma.

Proof of the Theorem. We observe, first of all, that if two functions
fo and fl satisfy condition B) of the proposition in § 1.3 for the point zo,
then they satisfy condition A) for arbitrary points z', z", z' -# z", in a
sufficiently small neighbourhood of zoo .
F or the function
is analytic and
F(z, z) = fo(z) f{(z) - fl(z) f~(z).
°
Therefore F(zo, zo) -# 0, hence F(z 1, Z2) -# for points z 1 and Z2 suf-
ficiently close to zo, from which our assertion follows.
Obviously, if condition A) or B) holds for certain functions and points
z', z" or zo, then it also holds for sufficiently close points. Using the
lemma we choose a finite covering of the compact manifold X = DIG
by open sets Vi such that in Vi condition B) is satisfied for functions
{fO,i' f1,J, i = 1, ... , N. According to the remark made above there
exists a neighbourhood V of the diagonal in X x X such that at every
point of this set some pair fo,i' f1,i satisfies condition A). Since X x X - V
is a compact set, we can find a finite set of pairs of functions {fo,i' fl,d,
i = 1, ... , N, such that at any point of the space X x X some pair of
functions satisfies condition A).
Let the weight of the functions fo,i and f1,i be mi> and M = II mi,
ti = Mimi' We consider the system consisting of all products of the form
fi.li,f/li, (fo.;, fd 1i ,fJ:il f1:7\ i = 1, ... , N. Clearly all these are auto-
morphic forms of weight 2M. We show that for them conditions A)
and B) in the proposition of § 1.3 hold. Indeed, if the functions fo,i' and
f1,i satisfy condition A) at the points Zl, z", then the minor
rl'·fl,.)
(JO,l (Zl) (r l,-:-1fl,+1) (z") - (rl"f l,.) (z") (r l,-:-1fli:t"1(zl)
1,l JO,l 1,1 JO,l 1,1 JO,l 1,l
= fJ:i 1(Zl) fJ:i 1(z") fb(ZI) ff:i(Z") (fo,i(Z') f1,i(Z") - fo)z") fl,i(Z'))
is different from O. Condition B) is verified similarly, and this proves
the theorem.
Note. In the proof of the theorem we have made very little use of
specific properties of the interior of the unit disc D. Even the fact
that it is one-dimensional does not play an essential role in it. It can
be carried over almost without change to the case when D is any bounded
domain in (Cn and G a group acting discretely and freely on it for which
the factor space DIG is compact. Only we must understand by jg in the
definition of an automorphic form the Jacobian of the transformation
gEG.
3. Algebraic Curves and Automorphic Functions. In § 1 and § 2 we

have proved that algebraic curves of elliptic and parabolic type coincide
with the curves of genus 0 and 1. Therefore curves of hyperbolic type
must be those of genus g> 1. The theorem shows that these curves
coincide with compact manifolds of the form DIG, where D is the interior
of the unit disc and G a group acting discretely and freely on it.
Now we shall describe algebraically an embedding ofa curve X ~ DIG
in a projective space, which is defined by automorphic forms. Let f(z) be
an automorphic form of weight /'. The expression 1] = f(z) (dzt deter-
mines a holomorphic differential form of weight /' on D (the definition
of a holomorphic differential form is given in Exercise 4 to Ch. VIII, § 2,
that of a differential form of weight> 1 in Exercise 7 to Ch. III, § 5). From
the definition of an automorphic form it follows that 1] is invariant under
the automorphisms of G. Indeed,
g*1] = g*(f) (dg(z))l' = f -j~'j;l' (dzt = 1] •
Therefore 1] = n*w, where n is the projection D~ DIG, and w is a holo-

morphic differential form of weight l' on DIG. Finally, if qJ: DIG~X is
an isomorphism with the algebraic curve, then Wi = (qJ -1)* W is a holo-
morphic differential form of weight l' on X. Hence it follows that Wi is a
rational form of X. It is sufficient to take its quotient with any rational
form of the same weight on X; by Theorem 1 of Ch. VIII, § 3, this is a
rational function on X. The lemma in Ch. VIII, § 3, shows that Wi is a
regular form of weight [' on X.
It is easy to show that, conversely, every regular differential form of
weight [' on X is obtained in this way.
So we see that the space of automorphic forms of weight If is iso-

morphic to the space of regular differential forms of weight 1'.
Thus, the mapping by means of all automorphic forms of weight I' > 2
coincides with the mapping corresponding to the class If. Kx. In Ch. III,
§ 5.6, we have deduced from the Riemann-Roch theorem that this
mapping is an embedding for If ~ 3. Hence this is also true for the mapping
defined by' the automorphic forms of weight If. In addition, we obtain an
interesting analytic application of the Riemann-Roch theorem: the
dimension of the space of automorphic forms of weight If is finite and
(by the Riemann-Roch theorem) equal to
I(lf. Kx) = (21' - 1) (gx - 1).
As in § 2.3, it follows from Theorem 1 that the field <C(X) is iso-

morphic to the field of meromorphic functions on D that are invariant
under the group G. Such functions are called automorphic. Thus, every
curve of genus g > 1 can be uniformized by automorphic functions.
Let us compare the picture we have obtained with that which holds in
the parabolic case. In both cases the description of the curves reduces
to a description of certain discrete groups. In the parabolic case the
corresponding discrete groups are very simple: they are lattices in <c 1 .
What happens in the hyperbolic case?
Poincare found a general method of constructing groups that act
discretely and freely on the interior of the unit disc. His method is
based on the fact that a metric can be defmed in D in which analytic mo-
tions coincide with analytic automorphisms of D, and as a metric space D
is isomorphic to the Lobachevskii plane. In this metric isomorphic lines
of the Lobachevskii geometry correspond to circular arcs contained in
D and orthogonal to the unit disc, the boundary of D. We do not need
the definition of this metric. We only mention that in it the magnitude
of an angle is the same is that between circles (that is, between their
tangents at the point of intersection) in the Euclidean metric of the
plane <C 1 of a complex variable containing D.
Fig. 19
Poincare showed that every group G acting freely and discretely in D

for which the factor space DIG is compact is defined by a certain polygon
in the geometry described above. This polygon plays in the hyperbolic
case the same role as the fundamental parallelogram of the lattice Q
in the parabolic case and is called the fundamental polygon of G.
If the genus of an algebraic curve DIG is g, then the fundamental
polygon of G has 4g sides. We denote its sides in some direction of going
around the polygon by ai' a2' a'l, a~, a3' a4' a3' a~, ... , a2g-1, a~g-l' a~g,
in the order of the chosen direction of going around the polygon. Fig. 19
illustrates the case g = 2. Then the following relations hold:
1) the sides ai and ai are equal to each other.
2) the sum of the (interior) angles of the polygon is 2 n.
The group G is determined by its fundamental polygon in the fol-
lowing way. We denote by gi(i = 1, ... , 2g) the motions (without change
of orientation) that carry the side ai into ai with the opposite direction.
Then the motions gi generate G.
Conversely, if a polygon cP is given satisfying the conditions 1) and 2),
then the group G generated by the transformations gi acts on D discretely
and freely, and cP is its fundamental polygon. Geometrically this is
expressed by the fact that if F is the interior of cP, then by applying the
transformations gi first to F and then to the domains giF etc. we cover the
whole of D by domains that intersect only in sides of the boundary.
It would be natural to try and obtain from this picture a description
of classes of isomorphic curves of genus g> 1, by analogy with the
explanations at the end of § 2.3. However, the situation here is vastly
more complicated and less well investigated. The corresponding complex
space and even the algebraic variety can be defined precisely (in [4] the
reader can find an analytic, and in [23] an algebraic definition). It is
called the variety of moduli of curves of genus g. However, very little
is known of its properties. Of the problems stated in precise terms the
most interesting is the following: is the variety of moduli rational or
perhaps unirational? The rationality has been proved only for g = 2. The
problem of unirationality is apparently easier. For small values of the
genus (g = 3, 4, 5) it is not hard to prove. Severi has proved the uni-
rationality of the variety of moduli of curves of genus g for g< 11. Nothing
more is known.
Exercises
1. Show that the automorphic forms of weight 3 determine a projective embedding
of a compact manifold DIG.
2. Show that for a fIxed group G of automorphisms of the interior of the unit disc D
for which DIG is compact the equation I g*(h)j~ = 0 has infInitely many linearly inde-
geG
pendent solutions among functions h that are holomorphic and bounded in D.
§ 4. On the Uniformization of Manifolds of Large Dimension 401
3. Show that the genus g of a curve DIG and the area (in the sence of Lobachevskii
geometry) S of the fundamental polygon .p of a group G are connected by the relation
g - 1 = _1_ S. Hint: Use the theorem of Lobachevskii geometry according to which the
4n
sum of the angles of a triangle is less than n by the value of its area, and use also the con-
nection of the genus with the Euler characteristic.
§ 4. On the Uniformization of Manifolds of Large Dimension

1. Simple Connectivity of Complete Intersections. Hardly anything is
known on universal coverings and fundamental groups of varieties of
dimension greater than 1. We give some simple examples and make a few
remarks, with the object of throwing light on the nature of the problems
arising.
The basic new phenomenon that we encounter here is the following.
Among the smooth complete algebraic curves only one, the projective
line, is simply-connected, so that the transition to the universal covering
almost always reduces the study of the curve to that of another variety,
which one might hope, and which turns out, to be simpler. No such thing
happens for varieties of dimension ~ 2: among them very many are
simply-connected, and for them the transition to the universal covering
does not yield anything new. To make the expression "very many" a little
more precise we prove for one wide class of varieties of dimension ~ 2
that they are simply-connected; among them, in particular, all smooth
projective hypersurfaces.
Defmition. A projective variety Xc JPN of codimension n is called
a complete intersection if it is the intersection of n hypersurfaces that are
transversal at each of their points of intersection.
By our definition complete intersections are smooth algebraic varieties.
From now on we consider them over the field of complex numbers.
We show that if the dimension of a complete intersection X is greater
than 1, then the topological manifold X(<C) is simply-connected.
This is a consequence of a general result, which we use repeatedly
later:
Proposition. If V is an n-dimensional projective variety over the
field of complex numbers and W a hyperplane section of it for which
V - W is smooth, then the embedding W(<C) --+ V(<C) determines an iso-
morphism of their homotopy groups
in dimensions r < n - 1.
The proposition is a simple application of Morse theory (see, for

example, [21]). We have to make use of Theorem 7.4 in [21] and the
exact homotopy sequence of the pair (V(<C), W(<C»).
We prove simple connectivity of complete intersections by induction
on their codimension in projective space. At the first step of the induction
we must use the fact that the space lPn(<c) is simply-connected.
Let X C lP nbe the intersection ofn transversal hypersurfacesE l , ... , En'
We set Y=Eln .. ·nEn - l . Clearly Y is a complete intersection, and
by the inductive hypothesis the space Y(<C) is simply-connected. We
consider the Veronese mapping
vm:lP n -+lP M , M= (n+m)

m -1,
where m = deg En. Let V = vm(Y), W = vm(X). Evidently W is the section of

V by the hyperplane vm(En). Since the space V(<C) is homeomorphic to
Y(<C), it is simply-connected. We may apply the proposition and find
that for dim V> 2, that is, dim X ~ 2
Since the space X(<C) is homeomorphic to W(<C), it is also simply-

connected.
2. Example of a Variety with a Preassigned Finite Fundamental Group.

However, there exist many non simply-connected algebraic varieties of
any given dimension. Presently we shall give an illustration of this
phenomenon, which in a certain sense is opposite to what we talked about
in § 4.1. Namely, we show that for every finite group r and every integer
n ~ 2 there exists an n-dimensional complete algebraic variety whose
fundamental group is isomorphic to r.
To begin with, we construct the example for the case when = Sm isr
the symmetric group of degree m. To do this we consider the product
II = lPs x ... x lPs of m copies of the s-dimensional projective space. We
write a point x E II in the form x = (Xl'''', x m), Xi E ]PS.
On this variety the group Sm acts by permuting the individual points
The main step in the construction of the example is the construction of

the factor space II' = II/Sm, that is, of the normal variety II' and a finite
morphism qJ: II'-4 II', such that qJ(x) = qJ(x') if and only .if x' = go X
for some g E Sm.
Suppose that in the j-th copy of JrS homogeneous coordinates are

denoted by XO,j'"'' Xs,j' We introduce s + 1 auxiliary variables to, ... , ts'
and we consider the form
(1)
s
Lixj ' t) = L XiJi' X
i=O
j = (XO,j"'" xs,) .
We denote a monomial of degree m in the variables to, ... , ts by T(~),

and the number of them by N + 1. Then
N
F(x, t) = L F~(x) T(a) ,
a=O
where the F~(x) are forms in the variables Xi,j, linear in each system
XO,j"'" Xs,j' We consider the rational mapping
<p(X) = {F~(x), rx = 0, ... , N},
defined by these forms. This mapping is regular: if all the FaCi') = for
°
some point x E TI, then F(x, t) = 0, and this means that LiXj, t) = for some
°
j, that is, all the coordina tes of x j are 0. There is a simple connection between
<p and that embedding of TI as a closed subset fI of some projective space
which was constructed in Ch. I, § 5.1. Namely, it is easy to verify that
<p is a projection of fI and that for this projection the conditions of
Theorem 8 in Ch. I, § 5, hold, so that the mapping <p: TI--- TI' C IPN,
TI' = <p(TI) is finite.
Under the action of Sm on TI the factors in (1) are permuted, from
which it follows that <P' 9 = <p, that is, if x = g(y), then <p(x) = <p(y). Con-
versely, if <p(x) = <p(y), then x = g(y) for some 9 E G. For if <p(x) = <p(y),
then F(x, t) = cF(y, t), c # 0, and from the unique factorization of poly-
nomials it follows that the points Yj, j = 1, ... , m, are obtained by a
permutation of Xl'"'' Xm:
(2)
Now we show that TI' is a normal variety. For this purpose we
observe that the polynomials Fa(x l , ... , Xm) do not change when the
points Xl'"'' Xm are permuted. The converse is also true: every poly-
nomial in homogeneous coordinates of Xl'"'' xm that is homogeneous
in the coordinates of each of these points and invariant under arbitrary
permutations of the points is a form in the polynomials Fa. This is an
analogue to the fundamental theorem on symmetric functions and is
proved in exactly the same way. A proof can be found in older textbooks
of algebra, for example, [6J, Ch. XIX, § 89.
Let H be a form in homogeneous coordinates of points on II', and
Yell' the affine open set defined by the condition H"#O, X=cp-I(y);
it is also affine and is defined by the condition cp* H "# O. We verify that
the ring cp*k[YJ consists precisely of the elements of k[X] that are
invariant under Sm:
cp*k[YJ = k[XJSm. (3)
For a function f E k[XJ has the form
where HI is a form whose degree is equal to that of (cp* H)l. If f is in-

variant under Sm' then so is HI (the form cp* H is necessarily invariant).
Therefore it follows from the generalized theorem on symmetric functions
quoted above that Hl is a form in the polynomials Fa., and this means
that f E cp*k[Y].
From (3) it follows that k[YJ is integrally closed. For if a function
f E k(Y) is integral over keY], then cp*(f) is a fortiori integral over k[X].
But X is smooth and hence normal. Therefore cp*(f) E k[X]. When we
now apply (3), we find that cp*(f) E cp* k[ y], that is, f E k[ Y]. So we have
shown that Y is a normal variety. Since the hyperplane H = 0 was chosen
arbitrarily, it now follows that II' is normal.
Clearly from (3) an analogous equality follows for the field of frac-
tions cp* k(II) = k(II)Sm. From the simplest results of Galois theory it
now follows that k(II)/k(II') is a Galois extension with Galois group Sm.
In particular,
degcp =.m! (4)
We denote by LI e II the closed set consisting ofthose points (Xl' ... ' Xm)
for which Xi=X j for some i"#j, and we set LI'=cp(LI)eII', W=II-Ll,
w' = II' - LI'. If X' E W', then by (2), cp-l(X') consists of m! distinct points.
Comparing this with (4) we see that cp : W -> W' is an unramified mapping.
Since W is smooth, it follows from Corollary 3 to Theorem 8 in Ch. II,
§ 5, that W' is also smooth.
We have now constructed two smooth varieties Wand W' and an
unramified covering cp : W -> W' with automorphism group Sm. However,
this is not what we need, because both our varieties are incomplete.
To avoid this difficulty we intersect II' with a linear subspace L e IPn
such that L does not intersect LI' and that the variety Y = Ln II' is smooth.
Such a subspace exists and can be given by d independent linear equations

provided that
d> codimu,LI' . (5)
The variety consisting of the points (Xl"'" X",) e II for which xp = Xq
is of codimension s in II, hence
codimuLl = codimn,LI' = s,
so that condition (5) takes the form d> s. We can choose the linear space
so that the dimension of Y is determined by the theorem on the dimension
of an intersection:
dim Y=dim II' -d=m·s-d.
Obviously, by choosing s sufficiently large we can achieve that the
relations
dim Y=m's-d=n, d>s,
hold. For this it is sufficient that
s(m-1»n.
Since YnLl'=0, that is, YeW', we see that xe<p-l(y) is an un-

ramified covering of Y with automorphism group Sm.
All the preceding arguments were purely algebraic. Let us suppose
now that all the varieties are defined over the field of complex numbers <C.
As we have seen in Ch. II, § 5.3, the mapping of topological spaces
<p : X(<C) -+ Y(<C) is an unramified covering. The proposition in § 4.1
applied for r = 0 shows that X(<C) is connected. For X is obtained from II
by intersections with hypersurfaces that can be regarded as hyperplanes
under the standard embedding of II in a projective space. The same
proposition can be applied for r = 1. Since II(<C) is simply-connected, the
proposition shows that X(<C) is also simply-connected.
So we see that X(<C) is the universal covering of the manifold Y(<C) and
that Sm = 1tl(Y(<C))' Observe that by construction Y(<C) is projective.
Starting out from the unramified covering
<p:X-+Y
with the group Sm it is easy to construct a covering with an arbitrary

finite group r. To do this we assume that r e Sm. We have seen that
the extension <C(X)j<C(Y) has the Galois group Sm' and by Galois theory, to
the subgroup r there corresponds a subfield K such that <C(X)) K) <C(Y)
and <C(X)jK has the Galois group r. We denote by Y the normalization of
Y in K. According to the general properties of normalizations, we have
morphisms
x -'4 Y -'4 Y,
with
tp.<p=<p.
From the general properties of finite morphisms it follows that iii and tp
are finite. We show that Y is smooth and iii unramified. For since
deg <p = deg iii deg tp and since the number of inverse images <p - 1 (y) for
a closed point y E Y is equal to deg <p, the number of inverse images
tp-l(y) is equal to degtp, and for }IE Y the number of inverse images
(iiJ}-l(y) is equal to degiii. Thus, <p and tp are unramified, and since Y is
smooth, by Corollary to Theorem 8 of Ch. II, § 5, Y is also smooth. So we
see that X(C) is a universal covering for Y(C) and that nl(Y(C))=T.
This completes our construction. Observe that from the unproved
theorem on the normalization of a projective variety being projective it
follows that the variety Y we have constructed is projective.
3. Notes. As a supplement to the example constructed above it should

be noted that we can construct examples of a large number of projective
varieties whose fundamental groups are infmite. Thus, if X is an n-dimen-
sional Abelian variety, then by the results of Ch. VII, § 1.3, the manifold
X(<C) is homeomorphic to a 2n-dimensional torus and nl (X (<C)) = 7l 2n •
If n ;;;. 3, then according to the proposition in § 4.1 a smooth hyperplane
section Y of the torus X has the same fundamental group. Its universal
covering is a subvariety of <c n , but apparently nothing is known about
its structure.
In these examples we have encountered two types of fundamental
groups and of universal coverings of complete algebraic varieties.
Type I. The universal group n 1 (X) is finite. In that case it can be
shown that the fundamental covering is a complete algebraic variety,
and if X is projective, also a projective variety. In the case of dimension 1
this type is represented only by the line IP 1, the only variety of elliptic
type.
To define Type II in precise terms is difficult. However, in this case
the fundamental group is necessarily infinite, and the universal covering
is a "very large" analytic manifold, far removed from being projective or
complete algebraic. In the case of dimension 1 these are curves of para-
bolic and hyperbolic type. For dimension;;;. 2 they are Abelian varieties
or (by the remark at the end of § 3.2) varieties of the form DIG, where D
is a bounded domain in <cn and G a group acting on D discretely and
freely such that DIG is compact. Hyperplane sections of these varieties
also belong to this type.
In trying to characterize more accurately the second type of variety

we are led to the definitions of complex spaces of two types that playa
fundamental role in the general theory of complex spaces and manifolds.
A complex space X is said to be holomorphically convex if for every
sequence of points Xn E X having no limit points on X there exists a
function f holomorphic on X such that If(xn)l--+oo as n--+oo.
Every compact space is trivially holomorphically convex. Another
example are the spaces Xao ' where X is an algebraic affine variety. If
X CJAn, then even among the coordinates in JAn we can find a function
such as required in the definition.
A holomorphically convex analytic space X is said to be holo-
morphically complete if the holomorphic functions on it separate its
points, that is, for any two points x', x" E X, x' # x", there exists a holo-
morphic function f on X such that f(x') # f(x").
From Theorem 2 in Ch. VIII, § 2, it follows that a compact analytic
manifold is holomorphically complete only if it consists of a single
point. The same is true for compact analytic spaces.
Analytic spaces of the form Xao ' where X is an affine algebraic variety,
are obviously holomorphically complete. Altogether, holomorphically
complete analytic spaces occupy in the general theory of analytic spaces
a place analogous to affine varieties in algebraic geometry. For example,
they too are "antipodal" to compact spaces, like affine varieties are to
projective or complete varieties.
Now we can give a more accurate description of the two types of
universal coverings of algebraic varieties that occur in our examples.
In Type I the universal coverings are compact, and in Type II they are
holomorphically complete. It is natural to expect that the general case is
in some sense "a mixture" of these two extremes. In the theory of analytic
spaces there is a fundamental result that can be regarded as a more
precise definition of the term "mixture". This is the so-called reduction
theorem of Remmert, which asserts that every holomorphically convex
normal analytic space X has a proper holomorphic mapping f: X --+ y
onto a holomorphically complete space Y (a mapping f is said to be
proper if the inverse image of every compact set is compact; in particular,
its fibres are compact).
In view of this theorem an interesting question arises: is perhaps
the universal covering of a complete algebraic variety holomorphically
convex? (Perhaps it would be more cautious to confine ourselves to
projective varieties.)
Clearly all compact varieties are holomorphically convex. A typical
example of a variety that is not holomorphically convex is ([:2 - O. For
it can be shown that a function holomorphic on ([:2 - 0 is also holo-
morphic at 0 (analogous to the fact that a rational function on a smooth
algebraic variety is non-regular at the points of an integral divisor).

Therefore, if Xn ~ 0, then for every function j holomorphic on <e 2 - 0
we have j(xn)~ j(O). In fact, the variety <e 2 -0 is the universal covering
of non-algebraic varieties, for example, of Hopf surfaces (Ch. VIII,
§ 1.2). Kodaira has shown that no compact manifold whose universal
covering coincides with <e 2 - 0 is algebraic.
In § 3.2 we have remarked that if a bounded domain DC <en is the
universal covering of a compact manifold, then the latter is an algebraic
variety. On the other hand, it has been shown that every such domain D
is holomorphically convex. These and some other examples give reason to
hope that the answer to the question raised above is in the affirmative.
A proof of this would be a great advance in the problem of the structure
of universal coverings of algebraic varieties. At present almost nothing is
known on universal covering varieties of arbitrary algebraic varieties.
Apparently the only result in this direction is due to Griffiths, who has
shown that any projective variety contains an affine open subset whose
universal covering is isomorphic to a bounded domain in <en.
Exercises
1. Let C ClP2 be a smooth projective plane curve given by the equation F(x o, X" x 2 ) = 0
of degree n, V C 1P3 a projective surface given by the equation F(x o, X" x 2 ) = x 3, and
f: V - t 1P2 the projection with centre at (0 : 0 : 0: 1). Show that f: V - f - I (C) - t 1P2 - C is
an unramified covering and that V - f - I (C) is the universal covering for 1P2 - C. Deduce
that 1l: 1 (1P2- C)""Z/n.
2. Show that (1P')n/Sn = IPn.
3. Let X be a smooth projective curve, G = {1, g}, g(x, x') = (x', x) E X X X. Show that
the ringed space Y = (X x X)/G is a complex space and even a variety (notwithstanding
the fact that g has fixed points). Show that 1l:,(Y) "" HI (X).
Bibliography
[lJ Abhyankar, S. S.: Local analytic geometry. New York-London: Academic Press, 1964.
[2J Aleksandrov,P.S., Efremovich,V.A.: A survey of the fundamental concepts of
topology. Moscow: ONTI 1936.
[3J Algebraic surfaces. Trudy Mat. Inst. Steklov 75 (1965). English translation: Algebraic
Surfaces (by the members of the seminar of I.R.Safarevic). Amer. Math. Soc. Transla-
tions 75 (1967).
[4J Ahlfors, L. V., Bers, L.: Spaces of Riemann surfaces and quasiconformal mappings.
Moscow: Izdat. Inost. Lit., 1961. (Translation into Russian of two papers by Ahlfors
and four by Bers, see MR 24 A229.)
[5J Artin,M.: Algebraic spaces. Mimeographed notes. New Haven, Conn.; Yale
University, 1969.
[6J B6cher,M.: Introduction to higher algebra. Cambridge Mass.: Harvard Univ., 1909.
[7J Borevich,Z.I., Shafarevich,I.R.: Number theory. Moscow: Nauka, 1964. Translation,
New York and London: Academic Press, 1966.
[8J Bourbaki, N.: Elements of mathematics, Book III, General topology, 2 vols. Paris,
Reading, Mass.: Hermann Cie.-Addison-Wesley Publ. Co., 1966.
[9J Cartan,H.: Elementary theory of analyti9 functions of one or several complex
variables. Paris-Reading, Mass.: Hermann & Cie.-Addison Wesley Publ. Co., 1963.
[lOJ Cartan,H., Eilenberg,S.: Homological algebra. Princeton, N. 1.: Univ. Press, 1956.
[llJ Chern,S.S.: Complex manifolds. Univ. Recife 1959.
[12J Eilenberg,S., Steenrod,N.E.: Foundations of algebraic topology. Princeton, N. 1.;
Univ. Press, 1962.
[13J Fam,F.: Introduction to the topological investigation of Landau singularieties.
Moscow: Mir, 1970.
[14J Grothendieck,A.: Cohomologie locale des faisceaux coherents et theonemes de
Lefschetz locaux et globaux. 2 fasc., third rev. ed., Paris: Inst. Hautes Et. Sci., 1968.
[15J Grothendieck,A., Dieudonne,1.: Elements de geometrie algebrique, vol. 1. Berlin-
Heidelberg-New York: Springer-Verlag, 1971.
[16J Goursat,E.: A course in mathematical analysis, vol. 1. New York: Dover reprint, 1959.
[17J Gunning, R. c., Rossi, H.: Analytic functions of several complex variables. Englewood
Cliffs, N. 1.: Prentice-Hall, 1965.
[18J Hodge, W. V. D., Pedoe, D.: Methods of algebraic geometry, vol. II. Cambridge: Univ.
Press, 1952.
[19J Husemoller, D.: Fibre bundles. New York-London: McGraw-Hill, 1966.
[20J Lang, S.: Introduction to algebraic geometry. New York-London: Interscience, 1958.
[21J Milnor,l.W.: Morse theory. Princeton, N. 1.: Univ. Press, 1963.
[22J Moishezon, B. G.: An algebraic analogue to compact complex spaces with a sufficiently
large field of merom orphic functions. Izv. Akad. Nauk SSSR Ser. Mat. 33, 174-238,
323-367, 506-548 (1969).
[23J Mumford,D.: Geometric invariant theory. Berlin-Heidelberg-New York: Springer-
Verlag, 1965.
410 Bibliography
[24J Mumford, D.: Lectures on curves on an algebraic surface. Princeton, N. J.; Univ.
Press, 1966.
[25J Pontryagin,L.S.: Topologische Gruppen, second ed. 2 vols. Leipzig: Teubner, 1957.
[26J de Rham,G.: Varietes differentiables. Paris: Hermann&Cie. 1955.
[27J Samuel, P.: Methodes d'algebre abstraite en geometrie algebrique. Berlin-Heidelberg-
New York: Springer-Verlag, 1967.
[28J Seifert, H., Threllfall, W.: Lehrbuch der Topologie. Leipzig: Teubner, 1934.
[29J Serre,J.P.: Lecture Notes in Math. Vol. II, Algebre Locale multiplicites. Third ed.
Berlin-Heidelberg-New York: Springer-Verlag, 1975.
[30J Siegel,C.L.: Analytic functions of several complex variables. Princeton, N. J.: Inst.
for Adv. Studies, 1950.
[31J Springer, G.: Introduction to Riemann surfaces. Reading, Mass.: Addison-Wesley
Pub!. Co., 1957.
[32J Vinogradov,l.M.: Elements of number theory, seventh ed. Moscow: Nauka, 1965.
Translations: New York: Dover, 1954 or Oxford: Pergamon 1952.
[33J van der Waerden,B.L.: Algebra, 2 vols., fifth ed. Berlin-Gottingen-Heidelberg:
Springer-Verlag, 1960.
[34J Walker,R.J.: Algebraic curves, Princeton, N. J.; Univ. Press., 1950.
[35J Wallace,A.H.: Differential topology: First steps, New York-Amsterdam: W.A.Ben-
jamin, 1968.
[36J Weil,A.: Introduction it l'etude des varietes Kiihleriennes. Paris: Hermann & Cie.,
1958.
[37J Zariski,O., Samuel, P.: Commutative algebra, 2 vols. Princeton, N. J.; Van Nostrand
and Co. 1958/60.
Historical Sketch
This sketch does not pretend to give a systematic account of the history
of algebraic geometry. Its aim is to describe in very general terms how
the ideas and concepts with which the reader has become acquainted
in this book were created. In explaining the research of one mathe-
matician or another we often omit important work of his (sometimes
even the most important) if it has no bearing on the contents of our
book.
We shall try to formulate the results as closely as possible to the
way the authors have done it, using only occasionally contemporary
notation and terminology. In cases where this is not immediately obvious
we shall explain these investigations from the point of view of the concepts
and results of our book. Such places are marked by an asterisk * (at
the beginning and the end).
* Naturally algebraic geometry arose first as the theory of algebraic

curves. Only by going beyond the frame of rational curves do we
encounter properties of algebraic curves that are characteristic for
algebraic geometry. Therefore we leave aside the theory of conics,
which are all rational. Next in complexity and hence the first non-
trivial example are curves of genus 1, that is, elliptic curves and, in
particular, non-singular curves of the third degree. And historically the
first step in the development of the theory of algebraic curves consisted
in a clarification of the basic concepts and ideas of this theory in the
example of elliptic curves.
Thus, it would seem that these ideas developed in the same sequence in
which they are now set forth (for example, in Ch. I, § 1). However, in
one respect this is by no means the case. The complex of concepts and
results that we now call the theory of elliptic curves arose as part of
analysis, and not of geometry: as the theory of integrals of rational
functions on an elliptic curve. It was precisely these integrals that
originally were called by the name elliptic (they occur in connettion
with the computation of the arc length of an ellipse), and later the name
was transferred from them to functions and to curves. *
412 Historical Sketch
1. Elliptic Integrals. They were an object of study as early as the

XVII century, as an example of integrals that cannot be expressed in
terms of elementary functions and lead to new transcendental functions.
At the very end of the XVII century Jacob, and later Johann, Bernoulli
came up against a new interesting property of these integrals (see
J. Bernoulli [1J, Vol. 1, p. 252). In their investigations they considered
integrals expressing the arc length of certain curves. They found certain
transformations of one curve into another that preserve the arc length
of the curve, although the corresponding arcs cannot be superposed
to one another. It is clear that analytically this leads to the transformation
of one integral into another. In some cases there arise transformations
of an integral into itself. In the first half of the XVIII century many
examples of such transformations were found by Fagnano.
In general form the problem was raised and solved by Euler. He
communicated his first results in this direction in a letter to Goldbach in
1752. His investigations on elliptic integrals were published from 1756
to 1781 (see Euler [1 J).
Euler considers an arbitrary polynomial f(x) of degree 4 and asks
for the relations between x and y if
dx dy
(1)
Vf(x) = Vf(y) .
He regards this as a differential equation connecting x and y. The

required relation is the general integral of this equation. He finds this
relation: it turns out to be algebraic of degree 2 both in x and in y. Its
coefficients depend on the coefficients of the polynomial f(x) and on
one independent parameter c.
Euler formulates this result in another form: the sum of the integrals
CC
S 1~dx
an d Sfl 1~
dx.
IsequaI to a smg
. I·emtegraI:
o V f(x) 0 V f(x)
dx dx dx
! ! =!
CC fI Y
(2)
Vf(x) + Vf(x) Vf(x) ,
and '}' can be expressed rationally in terms of IX and p. Euler also brings
forward arguments why such a relation cannot hold if the degree of the
polynomial f(x) is greater than 4.
For arbitrary elliptic integrals of the form S
relation that generalizes (2):
( )d
f(x)
V.
x Euler proves a
S r(x) dx
V7W + I V7W - I V7W I V(y)dy,
CC fI r(x) dx Y r(x) dx 6
o = (3)
Historical Sketch 413
where y is the same rational function of 0( and p as in (2), and where ()

and V are also rational functions.
* The reason for the existence of an integral of the equation (1) and
of all its special cases discovered by Fagnano and Bernoulli is the pre-
sence of a group law on an elliptic curve with the equation S2 = f(t) and
the invariance of the everywhere regular differential form s-ldt under
translations by elements of the group. The relations found by Euler that
connect x and y in (1) can be written in the form
(x, Vf(x)) $ (c, Vf(c)) = (y, Vf(y)) ,

where $ denotes addition of points on the elliptic curve. Thus, these
results contain at once the group law on an elliptic curve and the existence
of an invariant differential form on this curve.
The relation (2) is also an immediate consequence of the invariance of
the form cp = I ~ • In it
V f(x)
and
(y, Vf(y)) = (0(, Vf(O())$(P, Vf(l3))
'" p '" y '" y y
JCP+Jcp=Jcp+J~cp=Jcp+Jcp=Jcp,
000", 0 '" 0
V
where tg is the translation by g = (0(, f(O()). Observe that we write
here the equation between integrals formally, without indicating the
paths of integration. Essentially this is an equation "to within a constant
of integration", that is, an equation between the corresponding differential
forms. This is how Euler understood them.
Finally, the meaning of the relation (3) will become clear later, in
connection with Abel's theorem (see 3.). *
2. Elliptic Functions. After Euler the theory of elliptic integrals

was developed mainly by Legendre. His investigations, beginning in 1786,
are collected in the three-volume "Traite des fonctions elliptiques et
des integrales Euleriennes (Legendre [1J).* In his preface to the first
supplement published in 1828 Legendre writes: "So far the geometers
have hardly taken part in investigations of this kind. But no sooner
had this book seen the light of day, no sooner had it become known to
scholars abroad, than I learned with astonishment as well as joy that two
young geometers, Herr Jacobi in Konigsberg and Herr Abel in Christiania,
have achieved in their works substantial progress in the highest branches
of this theory".
* Legendre called elliptic functions what we now call elliptic integrals. The contemporary
terminology became accepted after Jacobi.
Abel's papers on the theory of elliptic functions appeared in 1827-1829.

He starts out (see Abel [1], Vol. I, No. XVI, No. XXIV) from the elliptic
integral
where c and e are complex numbers; he regards it as a function of the

upper limit and introduces the inverse function A(8) and the function
LI(8) = V(1-
c2 A2)(1- e2..1 2). From the properties of elliptic integrals
known at that time [essentially, from Euler's relations (2) in 1.] he
deduces that the functions A(8 ± 8') and LI (8 ± 8') can be simply expressed
in the form of rational functions of A(8), A(8'), LI(8), and LI(8'). Abel
shows that both these functions have in the complex domain two periods
2w and 2w:*
lie dx
w-2
-
f V(1- C2 x 2 )(1 -
0 e2 x 2 )
He finds representations of the functions introduced by him in the form of
infinite products extended over their zeros.
As an immediate generalization of the problem with which Euler had
been occupied, Abel [1] (Vol. I, No. XIX) raises the question: "To list
all the cases in which the differential equation
dy = +a dx (1)
V(1- ciJ2)(1 - eiJ2) - V(1- c2x 2 )(1 - e2 x 2 )
can be satisfied by taking for y an algebraic function of x, rational or

irrational" .
This problem became known as the transformation problem for
elliptic functions. Abel showed that if the relation (1) can be satisfied by
means of an algebraic function y, then it can also be done by means of a
rational function. He showed that if c I = c, e 1 = e, then a must either
be rational or a number of the form Ii + -v=-Il, where fl' and fl are
rational numbers and fl > O. In the general case he showed that the
periods WI and WI of the integral of the left-hand side of (1), multiplied
by a common factor, must be expressible in the form of an integral linear
combination of the periods wand W ofthe integral of the right-hand side.
* As E. 1. Slavutin has remarked, already Euler [2J drew attention to the fact that the
Y dx
function S ~ has in the real domain a "modulus of multi-valued ness" similar to the
o i-x
inverse trigonometric functions.
Somewhat later than Abel, but independently, Jacobi [1J (Vol. I,

Nos. 3 and 4) also investigated the function inverse to the elliptic integral,
proved that it has two independent periods, and obtained a number of
results on the transformation problem. Transforming into series the
expressions for elliptic functions that Abel had found in the form of
products, Jacobi arrived at the concept of O-functions* and found
numerous applications for them, not only in the theory of elliptic
functions but also in number theory and in mechanics.
Finally, after Gauss's posthumous works were published, especially
his diaries, it became clear that long before Abel and Jacobi he had
mastered some of these ideas to a certain extent.
* The first part of Abel's results requires hardly any comment. The
mapping x = A(O), y = LI(O) determines a uniformization of the elliptic
curve y2 = (1 - c2 x 2 )( 1 - e2 x 2 ) by elliptic functions. Under the cor-
responding mapping f: the regular differential form cp = dx
([:1_ X
y
goes over into a regular differential form on ([:1 that is invariant under
translations by the vectors of the lattice 2w'll + 2m'll. This form differs
dx
by a constant factor from dO, and we may assume that dO = f* - ,
dx y
that is, 0= J-. y
The integration of the equation (1) has the following geometric mean-
ing. Let X and Xl be elliptic curves with the equations u2 = (1 - c2 x 2 )
. (1 - e2 x 2 ) and v2 = (1 - ci y2) (1 - ei y2). The point is to investigate curves
C CX X Xl (which corresponds to an algebraic relation between x and y).
Since an elliptic curve is its own Picard variety (see Ch. III, § 3.5), C deter-
mines a morphism f : X - Xl. This makes it clear why the problem reduces
to the case when y is a rational function of x. According to Theorem 4
in Ch. III, § 3, f can be regarded as a homomorphism of the algebraic
groups X and Xl. Thus, Abel studied the group Hom (X, Xl) and for
X = Xl the ring End X. A homomorphism f E Hom (X, Xl) deter-
mines a linear transformation of the one-dimensional spaces
f* :0 1[X 1]_01 [X], which is given by a single number, the factor ± a
in (1). See also Exercises 7, 8, and 9 to Ch. IX, § 2. *
3. Abelian Integrals. The transition to arbitrary algebraic curves

proceeded entirely within the framework of analysis: Abel showed that
the basic properties of elliptic integrals can be generalized to integrals of
arbitrary algebraic functions. These integrals later became known as
Abelian integrals.
* O-functions occured first in 1826 in a book by Fourier on the theory of heat.
In 1826 Abel wrote a paper (see Abel [1J, Vol. I, No. XII), which was
the beginning of the general theory of algebraic curves. He considers in
it an algebraic function Y determined by two equations
X(X,y)=o, (1)
and
O(x, y) =0, (2)
where O(x, y) is a polynomial that depends, apart from x and y, linearly
on some parameters a, d, ... , the number of which is denoted by cx. When
these parameters are changed, some simultaneous solutions of (1) and (2)
may not change. Let (Xl' Y1), ... , (X Il ' YIl ) be variable solutions, and
I(x, y) an arbitrary rational function. Abel shows that
Xl
f I(x, y) dx + ... + f I(x, y) dx = f V(g) dg ,

x~
(3)
o 0
where V(t) and g(x,y) are rational functions depending also on the param-
eters a, ai, .... Abel interpreted this result by saying that the left-hand
side of (3) is an elementary function.
Using the freedom in choosing the parameters a, ai, ... Abel shows
f I(x, y) dx can be expressed in
Xi
that the sum of any number of integrals

o
terms of j l - cx such integrals and a term of the same type as that on the
right-hand side of (3). He establishes that the number j l - cx depends only
on (i). For example, for y2 + p{x), where the polynomial p is of degree 2m,
we have jl-cx=m-1.
Next Abel investigates for what functions I the right-hand side of (i)
does not depend on the parameters a, d, .... He expresses I in the form
f{l (x'{), , and he
2 X, Y Xy
shows that 12 = 1, and 11 satisfies a number of
restrictions as a consequence of which the number y of linearly independ-
ent ones among the required functions I is finite. Abel shows that
y ~ jl- cx and that y = jl- cx, for example, if (using a much later terminolo-
gy) the curve X{x, y) = 0 has no singular points.
* The discussion of the solutions (Xl, Y1), ... , (x ll ' yJ of the system
consisting of (1) and (2) leads us at once to the contemporary concept of
equivalence of divisors. Namely: let X be the curve with the equation (1)
and D). the divisor cut out on it by the form 0). (in homogeneous co-
ordinates), where A. is the system of parameters a, d, .... By hypothesis,
D). = D;. + Do, where Do does not depend on A.. Therefore all the
D). = (Xl' Y1) + ... + (x ll ' YIl) are equivalent to each other. The problem
with which Abel was concerned reduces to the investigation of the
PI PI"
sum f cp + ... + f cp, where cp is a differential form on X, CXi and Pi are
a1 aJ.L
points on X, (OC1) + ... + (OC Il ) "'" (Pl) + ... + (P Il ). We give a sketch of a

proof of Abe1's theorem that is close in spirit to the original proof. We
may assume that
(OC 1) + ... + (oc ll ) - (Pl) - ... - (P Il ) = (g), g E <C(X),
(OC1) + ... + (ocJ = (g)o, (Pl) + ... + (PJ = (g)oo .
We consider a morphism g: X _]pl and the corresponding extension
<C(X)/<C(g). For simplicity we assume that this is a Galois extension
(the general case easily reduces to this), and we denote its Galois group
by G. The automorphisms (J' E G act on the curve X and the field <C(X)
and carry the points oc l , ... , ocllintoeach other, because {oc l , ... , ocll } =g-l(O).
Therefore {OC1, ... ,OCIl}={(J'OC,(J'EG}, where oc is one of the points OCi.
Similarly {Pl, ... ,PIl}={(J'P,(J'EG}. Representing qJ in the form udg,
we see that
Il Pi t1P P
i"fll qJ= t1~GL udg= !C~G (J'u)d g . (4)
The function v = L (J'U is contained in <C(g), and Abe1's theorem follows

from this. t1 E G Xi
I xj
J
We see that every sum of integrals ~ f(x, y) dx can be expressed as
, 0
a sum of I integrals L J f(x, y) dx + JV(g) dg if the equivalence
j=lO
I
L ((OCi) - 0) "'" L ((oc}) - 0) (5)
i j= 1
holds, where OCi = (Xi, Yi), OC} = (X)' Y}), and 0 is the point with X = O. From
the Riemann-Roch theorem it follows at once that the equivalence (5)
holds (for arbitrary OCi and certain oc) corresponding to them), with I = g
(see Exercise 19 to Ch. III, § 5). Thus, the constant fl- oc introduced
by Abel is the same as the genus.
If qJ E .0 1[X], then also v dgE .0 1[1Pl], where V= L (J'U in (4). Since
t1EG
.0 1[1Pl] = 0, in this case the term on the right-hand side of (3) dis-
appears. Hence it follows that y ~ g. In some cases arising naturally
the two numbers coincide.
We see that this paper of Abe1's contains the concept of the genus of
an algebraic curve and the equivalence of divisors and gives a criterion
for equivalence in terms of integrals. In the last relation it leads to
the theory of Jacobian varieties of algebraic curves (see § 5). *
4. Riemann Surfaces. In his dissertation published in 1851 Riemann

[1] (No.1) applied a completely new principle of investigating functions
of a complex variable. He assumes that the function is given not on the

plane of a complex variable but on some surface that "extends in many
sheets" over this surface. The real and imaginary parts of this function
satisfy the Laplace equation. This function is uniquely determined if
the points are known at which it becomes infinite, the curves along
which cuts make it single-valued, the character of its singularities at these
points, and the many-valuedness in passing through these curves.
Riemann also works out a method of constructing a function from
these data that is based an a variational principle which Riemann called
"the Dirichlet principle".
In the first part of the paper "Theory of Abelian functions", which
appeared in 1857, Riemann [1] (No. II) applied these ideas to the theory
of algebraic functions and their integrals. The paper begins with the
investigation of properties of the corresponding surfaces that belong, as
Riemann says, to Analysis Situs. By means of an even number 2p of
cuts the surface becomes a simply-connected domain. By arguments taken
from Analysis Situs he shows that p = w/2 - n + 1 where n is the number
of sheets and w the number of branch points of the surface over the plane
of the complex variable (taken with the appropriate multiplicities).
Riemann investigates functions that, speaking generally, are many-
valued on the surface, but single-valued in the domain obtained after
making the cuts, and on passing through the cuts their values change by
constants, the so-called moduli of periodicity of the function. The
Dirichlet principle gives a method of constructing such functions. In
particular, there are p linearly independent everywhere finite such
functions: the "integrals of the first kind". Similarly functions are con-
structed that become infinite at given points. In order to form from them
functions that are single-valued on the surface one has to equate to
zero their moduli of periodicity. Hence it follows that among the single-
valued functions that become infinite only at m given points not fewer
than m - p + 1 are linearly independent and, if m > p, this inequality
becomes an equality for points in "general" position.
Riemann shows that all the functions that are single-valued on a given
surface are rational functions of two of them: sand z, which are con-
nected by a relation F(s, z) = O. He calls two such relations belonging
to one "class" if they can be rationally transformed into one another.
In that case the corresponding surfaces have one and the same number p.
But the converse is not true. Studying the possible dispositions of the
branch points of surfaces, Riemann shows that the set of classes depends
for p > 1 on 3p - 3 independent parameters, which he calls "moduli".
* The surfaces introduced by Riemann closely correspond to the con-
temporary concept of a one-dimensional complex analytic manifold;
these are the sets on which analytic functions are defined. Riemann
raises and solves the problem of the connection of this concept with that
of an algebraic curve. (The appropriate result is called Riemann's
existence theorem.)
This circle of Riemann's ideas did not by any means become clear at
once. An important role in their clarification is played by Klein's lectures
[2], in which he emphasizes that a Riemann surface a priori is not
connected with an algebraic curve or an algebraic function. A definition
of a Riemann surface that differs only terminologically from the presently
accepted definition of a one-dimensional analytic manifold was given by
H. Weyl [1].
Riemann's paper marks the beginning of the topology of algebraic
curves. The topological meaning of the dimension p of the space Q1 [X]
is explained in it: it is half the dimension of the first homology group
of the space X(CC).
Analytically Riemann proves the inequality leD) ~ degD - p + 1. The
Riemann-Roch equality was then proved by his pupil Roch.
Finally, in this paper the field k(X) emerges for the first time as an
original object connected with a curve X, and the concept of a birational
isomorphism appears. *
5. The Inversion Problem. Already Abel had raised the question of

the inversion of integrals of arbitrary algebraic functions. He observed, in
particular, that the function inverse to the hyperelliptic integral con-
V
nected with 1p(x) has periods equal to half the value of this integral
taken between two roots of the polynomial 1p (see Abel [1], Vol. II,
No. VII).
Jacobi drew attention to the fact that we are concerned here with a
function of a single complex variable having more that two periods if the
integral is not elliptic, and that this is impossible for a reasonable function.
If X is a polynomial of degree 5 or 6, Jacobi proposes to consider the
pair of functions
x dx Y dx x xdx xdx
u=J-+J-,
Y
ov'X ov'X v= J lI'v + 0J lI'v'

o vX vX
He suggests expressing x - y and xy as analytic functions of the two
variables u and v and conjectures that this expression is possible by
means of a generalization of O-functions (see Jacobi [1], Vol. II, Nos. 2
and 4). This conjecture was verified in a paper by Gopel [1] published
in 1847.
The second part of Riemann's paper [1] (No. II) on Abelian functions
is concerned with the connection between O-functions and the inversion
problem in the general case. He considers a series in p variables

O(v) = Le F (m)+2(m,v) ,
( 1)
m
where m = (m1' ... , mp) ranges over all integral p-dimensional vectors,
v = (V1' ... , vp), (m, v) = ~miVi> F(m) = ~Ct.j/mjmz, Ct.jl = a/j. This series
converges for all values of v if the real part of the quadratic form F is
negative definite. The main property of the function 0 is the equality
O(v + nir) = O(v) , O(v + Ct. j ) = eLj(v)O(v) , (2)
where r is an integral vector, Ct. j a column of the matrix (Ct.jl), and Lj(v)
a linear form.
Riemann shows that one can choose cuts a1' ... , a p ' b 1, ... , b p which
make his surface simply-connected, and a basis U 1 , .•. , up of everywhere
finite integrals on this surface such that the integrals Uj over al are
equal to 0 for j+ I and to ni for j= I, and the same Uj over bl form a
symmetric matrix (Ct.j/), satisfying conditions under which the series (1)
converges. He considers the function 0 corresponding these coefficients
Ct.j / and the function O(u - e), where U = (U1' ... , up) (the Ui are everywhere
finite integrals) and e is an arbitrary vector.
Riemann shows that O(u - e) has on the surface p zeros 171' ... , 17p or
vanishes identically. For a suitable choice of the lower limits in the
integrals Ui in the first case
(3)
where the congruence is taken modulo integral linear combinations of the
periods of the integrals Ui. In this way the points 171' ... , 1'J p are uniquely
determined. In the second case there also exist points 171' ... , 17 p -2 such
that
(4)
Riemann knew that the periods of any 2n-periodic function of n
variables satisfy relations similar to those that are necessary for the
convergence of the series defining the O-function. These relations between
the periods were described explicitly by Frobenius [1J, who showed
that they are necessary and sufficient for the existence of non-trivial
functions satisfying the functional equation (2). Hence it follows that
these relations are necessary and sufficient for the existence of a mero-
morphic function with given periods that cannot be reduced by a linear
change of variables to a function of a smaller number of variables.
One only has to apply the theorem that every 2n-periodic analytic
function can be represented as a quotient of entire functions satisfying
the functional equation of the O-function. This theorem, stated by
Weierstrass, was proved by Poincare [2]. In 1921 Lefschetz [1J proved
that when the Frobenius relations hold, the O-functions determine an

embedding of the manifold (f;"/o' (where 0, is the lattice corresponding
to the given period matrix) into a projective space.
* The inversion problem is connected with questions which in this
book we only touched upon incidentally, often without proofs. The matter
concerns the construction of the J aco bian variety of an analytic curve and
properties of arbitrary Abelian varieties (see Ch. III, § 3.5, and Ch. VIII,
§ 1.3).
If 0 is a fixed point on a curve X, then f(x) = x - 0 is evidently an
algebraic family of divisors of degree zero on X. The basis of this family
coincides with X. By the definition of the Jacobian variety Jx of X (we
recall that this is the name for the Picard variety if X is a curve) there
exists a morphism cp: X -Jx , which is an embedding if the genus p
of X is different from O. It can be shown that cp*: 0,1 [Jx ] _ 0,1 [X] is
an isomorphism. Therefore in the representation
(5)
the 2p-dimensionallattice 0, C CP consists ofthe periods of p independent
differential forms OJ E 0,1 [X]. Riemann also starts out from this analytic
specification of the Jacobian variety and then develops an algebraic
method of investigating it.
If Do is an arbitrary effective divisor of degree p, then g(Xl' ... , xp)
= Xl + ... + xp - Do determines a family of divisors of degree 0 on X.
A basis of this family can be taken to be the factor space XP/S p of the
product of p copies of X with respect to the symmetric group acting by
permutations of the factors. By the definition of the Jacobian variety
there exists a morphism 1p: XP/S p - Jx . It follows easily from the Riemann-
Roch theorem that it is an epimorphism and one-to-one on an open set
in Jx . Therefore it is a birational isomorphism. In the analytical represen-
tation 1p takes the form (3), and by definition it is not one-to-one at those
points (Xl, ... , Xp) for which l(xl + ... + xp) > 1. It follows from the
Riemann-Roch theorem that this is equivalent to the condition
l(K - Xl - ... - xp) > 0, that is (because degK = 2g - 2), to
Xl + ... +xP"'~ - Yl_ - ..._- Yp -2
for certain points Yl' ... ' Yp-2. The latter relation is the same as (4), to
within the additional term K, hence to a shift by a point in Jx .
The Frobenius relations are the condition for the analytic manifold
eel/O, to be projective. They are written down in Ch. VIII, § 1.4 [formulae
(3) and (4)]. *
6. Geometry of Algebraic Curves. So far we have seen how the

concepts and results that nowadays form the foundation of the theory of
algebraic curves have been created under the influence and within the
framework of the analytic theory of algebraic functions and their
integrals. A purely geometric theory of algebraic curves developed
independently of this trend of research. For example, in a book published
in 1834 Plucker found formulae connecting the class, the degree of a
curve, and the number of its double points (see Exercise 13 to Ch. IV, § 3).
There he also proved the existence of nine points of inflexion on a plane
cubic curve (see Exercise 4 to Ch. VIII, § 1). But research of a similar kind
took second place in the mathematics of the time no deeper ideas were
linked with it.
Only in the period following the era of Riemann did the geometry of
algebraic curves occupy a central place in the contemporary mathematics,
alongside the theory of Abelian integrals and Abelian functions. Basically
this change of view point was connected with the name of Clebsch.
Whereas for Riemann the foundation is the function, Clebsch takes as the
fundamental object the algebraic curve. One can say that Riemann
considered a finite morphism f: X -4 IP 1, and Clebsch the algebraic
curve X itself. In the book by Clebsch and Gordan [1] a formula is
deduced for the number p of linearly independent integrals of the first
kind (that is, for the genus of the curve X), which expresses it in terms of
the degree ofthe curve and the number of singular points (see Exercise 12
to Ch. IV, § 3). There it is also shown that for p = 0 the curve has a rational
parameterization, and for p = 1 becomes a plane cubic curve.
An error Riemann had made turned out to be exceptionally useful for
the development of the algebraic-geometrical aspect of the theory of
algebraic curves. In the proof of his existence theorems he had regarded
as obvious the solubility of a certain variational problem: the "Dirichlet
principle". Before long Weierstrass showed that not every variational
problem has a solution. Therefore Riemann's results remained unfounded
for some time. One of the ways out was an algebraic proof of these
theorems: they were stated essentially in algebraic form. These investi-
gations, which were undertaken by Clebsch (see Clebsch and Gordan [1]),
furthered considerably the clarification of the essentially algebraic-
geometrical character of the results of Abel and Riemann, hidden under
an analytic cloak.
The trend of research begun by Clebsch achieved its bloom in the work
of his pupil M. Noether. Noether's ideas are particularly clearly out-
lined in his joint paper with Brill [1]. In it the problem is raised of
developing the geometry on an algebraic curve lying in a projective plane,
as the collection of results that are invariant under biunique (that is,
birational) transformations. The foundation is the concept of the group
of (coincident or distinct) points of the curve. They consider systems
of groups of points that cut out on the original curve linear systems of
curves (that is, systems whose equations form a linear space). It can
happen that all groups of such a system contain a common group G,
that is, consist of G and another group G'. The system of groups G'
obtained in this way is called linear. If the dimension of the linear (pro-
jective) space of equations of the cutting curves is equal to q and the
groups G' consists of Q points, then the system is denoted by g~). Two
groups of one and the same system are called corresidual. Clearly this
corresponds to the modern concept of equivalence of effective divisors,
and if G is contained in a linear system g~), then in the modern notation
deg G = Q, l( G) > q + 1 (we recall that l( G) is the dimension of the vector
space and q that of the corresponding projective space).
Every group of points determines a largest linear system g~) con-
taining all the groups corresidual to the given one. The numbers q and Q
are connected by the Riemann-Roch theorem, which is proved purely
algebraically.
Of course, the Riemann-Roch theorem presupposes a definition
analogous to the canonical class. It is given without appeal to the concept
of a differential form, but the connection with this concept is very easily
established. For if a curve of degree n has the equation F = 0 and is
smooth, then the differential forms W E Ql [X] can be written in the form
W = ;, dx, where <p is a homogeneous polynomial of degree n - 3
v
(Ch. III, § 5.4). It can be shown that if the curve has only the simplest
singularities, then this expression remains valid if it is required that <p
vanishes at all singular points. These polynomials are said to be associ-
ated. Associated polynomials of degree n - 3 determine the linear system
that is an analogue to the canonical class.
In their paper Brill and N oether consider a mapping of a curve into the
(p - i)-dimensional projective space defined by the associated polynomials
of degree n - 3. Its image is called a normal curve. They show that a
single-valued (in present-day terminology, birational) correspondence of
curves reduces to a projective transformation of normal curves (provided
that the curves are not hyperelliptic).
Noether [1] applied these ideas to the investigation of space curves.
In modern language we can say that his paper is concerned with the study
of the irreducible components of the Chow variety of curves in three-
dimensional space.
7. Many-Dimensional Geometry. At the beginning of the second

half of the nineteenth century many special properties of algebraic
varieties of dimension greater than 1, mainly surfaces, had been found. For
example, cubic surfaces had been investigated in detail, in particular,
Salmon and Cayley had proved in 1849 that on any cubic surface without
singular points there are 27 distinct lines. However, for a long time these
results were not combined by any general principles and were not
connected with the deep ideas that had been worked out at the time in the
theory of algebraic curves.
The decisive step in this direction was taken, apparently, by Clebsch.
In 1868 he published a small note [1] in which he considers algebraic
surfaces from the point of view (using modern terminology) of birational
isomorphism. He considers everywhere finite double integrals on the
surface and mentions that the maximal number of linearly independent
among them is invariant under birational isomorphism.
These ideas were developed in Noether's paper [2], which consists of
two parts. As is clear from the very title, in it he considers varieties of
an arbitrary number of dimensions. However, the major part of the
results refers to surfaces. This is typical for the whole subsequent period of
algebraic geometry: although very many results were in fact true for
varieties of arbitrary dimension, they were stated and proved only for
surfaces.
In the first part Noether considers "differential expressions" on a
variety of arbitrary dimension, and it is interesting that he writes down
an integral sign only once. Thus, here the algebraic character of the
concept of a differential form already becomes formally obvious. N oether
considers only forms of maximal degree. He shows that they make up a
finite-dimensional space whose dimension is invariant under single-
valued (that is, birational) transformations.
In the second part he considers curves on surfaces. (Only the last
section contains some interesting remarks on three-dimensional varieties.)
Noether gives a description of the canonical class (in modern terminology)
by means of associated surfaces, analogous to the way in which this was
done earlier for curves. He raises the question of the surfaces V that
cut out on a curve C lying on V its canonical class (again in modern
terminology). He calls the curves cutting them out on V associated with
C and gives an explicit description for them which leads him to a formula
for the genus of a curve on a surface. This formula essentially is the
same as (1) in Ch. IV, § 2.3; however, an understanding of the fact that
an associated curve is of the form K + C was achieved only 20 years later
in the work of Enriques.
In the same paper Noether investigates the concept of an exceptional
curve, which contracts to a point under a birational isomorphism.
The most brilliant development of the ideas of Clebsch and Noether
came not in Germany, but in Italy. The Italian school of algebraic
geometry exerted an immense influence on the development of this
branch of mathematics. Undoubtedly many ideas created by this school
have so far not been fully understood and developed. The founders ofthe
Italian geometrical school are Cremona, C. Segre, Bertini. Its most im-
portant representatives are Castelnuovo, Enriques, and Severi. Castel-
nuovo's papers began to appear at the end of the 1880's. Enriques was a
pupil (and relative) ofCastelnuovo. His papers appeared at the beginning
of the 1890's. Severi began to work about ten years after Castelnuovo and
Enriques.
One of the main achievements of the Italian school is the classifi-
cation of algebraic surfaces. As a first result we can quote here a paper of
Bertini [1] in which a classification of involutory transformations of the
plane is given. The matter concerns (in present-day terminology) the
classification, to within conjugacy in the group of birational aut om or-
phisms of the plane, of all group elements of order 2. The classification
turns out to be very simple, in particular, it is easy to derive from it
that the factor space of the plane with respect to a group of order 2 is a
rational surface. In other words, if a surface X is unirational and a
morphism f: IP2 ~ X is of degree 2, then X is rational.
The general case of Liiroth's problem for algebraic surfaces was
solved (affirmatively) by Castelnuovo [1]. After this he raised the problem
of characterizing rational surfaces by numerical invariants and solved
it in [2]. The classification of surfaces that we have explained in Ch. III,
§ 5.7, was obtained by Enriques in a series of papers, which was completed
in the first decade of our century (see Enriques [2]).
In the context of Liiroth's problem for three-dimensional varieties
Fano investigated certain types of these varieties, suggesting a proof of
the fact that they are not rational. Enriques had shown that many ofthem
are unirational. This would give a negative solution of Luroth's problem,
but Fano pointed out many obscure places in the proof. Some intermediate
propositions turned out to be not true. The problem was solved finally
when the last pages of this book were already written. V.A. Iskovskii
and Yu. I. Manin have shown that Fano's basic idea can be made to work.
They have shown that smooth hypersurfaces of degree 4 in IP4 are non-
rational (the fact that some of them are unirational had been proved by
B. Segre). Simultaneously Griffiths and Clemens have found a new
analytical method of proving that certain varieties are non-rational, for
example, smooth hyper surfaces of degree 3 in IP4 (see Exercise 18 to
Ch. III, § 5). Of course, these results are only the first steps on the way to a
classification of unirational varieties. *
The main tool of the Italian school was the investigation of families
of curves on surfaces -linear and algebraic (the latter were called
continuous). This led to the concept of linear and algebraic equivalence
(in our book linear equivalence is simply called equivalence). The con-
nection between these two concepts was first investigated by Castelnuovo
[3]. He discovered a link of this problem with an important invariant of
* A third method to construct non-rational but unirational threefolds was discovered by
Artin and Mumford. See v.A.lskovskii and Yu.I.Manin, Mat. Sborn. 86 (1971),140-166,
C.H.Clemens and P.Griffiths, Ann. of Math. 95 (1972) 281-358; M.Artin and D.Mumford
Proc. of Lond. Math. Soc. XXV (1972) 75-95. (Footnote to corrected printing, 1977).
the surface, the so-called irregularity. We do not give here the definition
of irregularity, which was used by Castelnuovo; it is connected with
ideas close to the cohomology theory of sheaves. Formula (1) below
gives another interpretation of this concept.
Castelnuovo [3J proved that if not every continuous system of curves
is contained in a linear system (that is, if algebraic and linear equivalence
are not one and the same thing), then the irregularity of the surface is
different from zero. Enriques [1J proves the converse proposition.
Furthermore, he shows that every sufficiently general curve (in an
exactly defined sense) lying on a surface of irregularity q is contained in an
algebraically complete (that is, maximal) continuous family, which is
stratified into linear families of the same dimension, and the basis of
the stratification is a variety of dimension q. Castelnuovo [1J showed
that the fibering of linear systems (that is, classes of divisors) determines
on a q-dimensional basis of the fibering constructed by Enriques a
group law by virtue of which this basis is an Abelian variety and is
therefore uniformized by Abelian (2q-periodic) functions. This Abelian
variety does not depend on the curve from which we have started out and
is determined by the surface itself. It is called the Picard variety of this
surface.
The irregularity turned out to be connected with the theory of one-
dimensional differential forms on the surface, the beginning of which goes
back to Picard [1 J; in this paper it is proved that the space of every-
where regular forms is finite-dimensional. In 1905 Severi and Castelnuovo
proved that this dimension is the same as the irregularity; in our notation
(1)
Severi [1J investigated the group of classes relative to algebraic
equivalence and proved that it is finitely generated. His proof is based
on a connection of the concept of algebraic equivalence with the theory of
one-dimensional differential forms. Namely, an algebraic equivalence
nlCl + ... +nrCr~Oisequivalenttothefactthatforsomeone-dimensional
differential form the set of its "logarithmic singularities" coincides
exactly with the curves C 1 , ... , Cr, taken with the multiplicities nl' ... , nr •
(A curve C is a logarithmic singularity of multiplicity n for a form OJ
iflocally OJ = nf- 1 df, where f is a local equation of C.) Picard had already
proved earlies that the so-defined relation of equivalence by means of
differential forms gives rise to a finitely generated group of classes (see
Picard and Simart [1J).
8. The Analytic Theory of Manifolds. Although a considerable part

of the concepts of algebraic geometry arose in analytic form, their
algebraic meaning cleared up in time. Now we pass on to concepts and
results which are essentially connected with analysis (at least from the
present point of view).
At the beginning of the 1880's there appeared papers by Klein and
Poincare devoted to the problem of uniformization of algebraic curves by
automorphic functions. The aim was to uniformize arbitrary curves by
functions that are now called automorphic, just as elliptic functions
uniformize curves of genus 1. (The term "automorphic" was proposed by
Klein, previously these functions were called by various names.) Klein
[1] (No. 84) started out from the theory of modular functions. The
field of modular functions is isomorphic to the field of rational functions,
but one can consider functions that are invariant under various subgroups
of the modular group and so obtain more complicated fields. In particular,
Klein considered functions that are automorphic relative to the group
consisting of all transformations z ~ az + db in which a, b, e, and dare
ez+
integers, ad - be = 1, and
(: :) == (~ ~) (mod 7) .
He proved that these functions uniformize the curve of genus 3 with the
equation xgx l + XIX 2 + x~xo = o. The fundamental polygon of this
group can be deformed so as to obtain new groups uniformizing new
curves of genus 3.
A similar train of ideas lay at the basis of the papers by Klein [1]
(No. 101-103) and Poincare [1], (p.92, 108, 169), but Poincare used for
the construction of automorphic functions the series that now bear his
name. They both conjectured correctly that every algebraic curve admits
a uniformization by the corresponding group and made substantial
progress towards a proof of this result. However, a complete proof was
not achieved at the time, but only in 1907 by Poincare (and independently
by Koebe). An important role was played by the fact that by this time
Poincare had investigated the concept offundamental group and universal
covering.
The topology of algebraic curves is very simple and was completely
studied by Riemann. In the investigation of the topology of algebraic
surfaces Picard developed a method that is based on a study of the fibres
of a morphism f: X ~ lPl. The point is to find out how the topology of
the fibre f-l(a) changes when the point a E lPl changes, in particular,
when this fibre acquires a singular point. He proved, for example, that
smooth surfaces in lP 3 are simply-connected (see Picard and Simart [1],
Vol. I). By this method Lefschetz [2], [3] obtained many deep results on
the topology of algebraic surfaces, and also on varieties of arbitrary
dimension.
The study of global properties of analytic manifolds began fairly

recently (Hopf [1J, A. Weil [2J). This domain developed vigorously in
the 1950's, in connection with the creation and application by Cartan and
Serre ofthe theory of analytic coherent sheaves (see Cartan [1J and Serre
[1J). We do not give a definition of this concept-it is an exact analogue
of the concept of an algebraic coherent sheaf (but we must emphasize
that the analytic concept was introduced before the algebraic one).
One of the basic results of this theory was the proof that the cohomology
groups (and, in particular, the groups of sections) of an analytic coherent
sheaf over a compact manifold are finite-dimensional. In this context
Cartan gave a definition of an analytic manifold that is based on the
concept of a sheaf and expressed the idea that the definition of various
types of manifolds is linked with the specification of sheaves of rings on
them.
9. Algebraic Varieties over an Arbitrary Field. Schemes. Formally

the study of varieties over an arbitrary field began only in the twentieth
century, but the foundations for this were laid earlier. An important role
was played here by two papers printed in one and the same issue ofCrelle's
Journal in 1882. Kronecker [1J investigates problems that would now-
adays be referred to the theory of rings of finite type without divisors of
zero and of characteristic O. In particular, for integrally closed rings he
constructs a theory of divisors.
The paper by Dedekind and Weber [1J is devoted to the theory of
algebraic curves. Its aim is to give a purely algebraic account of a
considerable part of this theory. The authors emphasize that they do not
use the concept of continuity anywhere, and their results remain true if
the field of complex numbers is replaced by the field of all algebraic
numbers.
The principal significance of the paper by Dedekind and Weber lies in
the fact that in it the basic object of study is the field of rational functions
on an algebraic curve. Concrete (affine) models are employed only as a
technical tool, and the authors use the term "invariance" to denote
concepts and results that do not depend on the choice of model. In this
paper the whole account becomes to a significant degree parallel to the
theory of fields of algebraic numbers. In particular, the analogy between
prime ideals of a field of algebraic numbers and points of the Riemann
surface of a field of algebraic functions is emphasized (we could say
that in both cases we are concerned with the maximal spectrum of a
one-dimensional scheme).
Interest in algebraic geometry over "non-classical" fields arose first
in connection with the theory of congruences, which can be interpreted as
equations over a finite field. In his lecture at the International Congress
of Mathematicians in 1908 Poincare says that the methods of the

theory of algebraic curves can be applied to the study of congruences in
two unknowns.
The ground for a systematic construction of algebraic geometry was
prepared by the general development of the theory of fields and rings in the
1910's and 20's.
In 1924 Artin published a paper (see Artin [1J, No.1) in which he
studied quadratic extensions of the field of rational functions of one
variable over a finite field of constants, based on their analogy to
quadratic extensions of the field of rational numbers. Of particular
importance for the subsequent development of algebraic geometry was
his introduction of the concept of the '-function of this field and the
formulation of an analogue to the Riemann hypothesis for the'-function.
We introduce (which Artin did not do) a hyperelliptic curve defined over
a finite field k for which the field in question is of the form k(X). Then the
Riemann hypothesis gives a best possible estimate for the number N
of points x E X that are defined over a given finite extension K/k, that
is, for which k(x) c K (just like the Riemann hypothesis for the field of
rational numbers gives a best possible estimate for the asymptotic
distribution of prime numbers). More accurately, the Riemann hypothesis
is equivalent to the inequality IN - (q + 1)1 < 2gvq, where q is the
number of elements of the field K and g the genus of the curve X.
Attempts to prove the Riemann hypothesis (which, as becomes clear at
once, can be formulated for any algebraic curve over a finite field) led
in the 1930's to work by Hasse and his pupils on the theory of algebraic
curves over an arbitrary field. Here the hypothesis itself was proved by
Hasse [1J for elliptic curves.
Strictly speaking, this theory concerns not curves but the correspond-
ing fields offunctions, and the authors nowhere use geometric terminology.
With this style one can become acquainted in the book by Hasse [2J
(see the sections devoted to function fields). The possibility of this bi-
rationally invariant theory of algebraic curves is connected with the
uniqueness of a smooth projective model of an algebraic curve. Therefore
great difficulties arise in applying this approach to the many-dimensional
case.
On the other hand, in a sequence of papers published under the
general title "Zur algebraischen Geometrie" in the Mathematische
Annalen between the end of the 1920's and the beginning of the 1930's,
van der Waerden made progress in the construction of algebraic geometry
over an arbitrary field. In particular, he set up a theory of intersections
(as we would say nowadays, he defined the ring of classes of cycles)
over a smooth projective variety.
In 1940 A. Weil succeeded in proving the Riemann hypothesis for an

arbitrary algebraic curve over a finite field. He found two ways of proving
it. One of them is based on the theory of correspondences of the curve X
(that is, divisors on the surface X x X), and the other on an analysis of its
Jacobian variety. Thus, in both cases many-dimensional varieties are
brought into play. The book by Weil [1] contains the construction of
algebraic geometry over an arbitrary field: the theory of divisors, cycles,
intersections. Here "abstract" (not necessarily quasiprojective) varieties
are defined for the first time by the process of pasting together affine
pieces (similar to Ch. V, § 3.2).
A definition of a variety based on the concept of a sheaf is contained
in the paper by Serre [2], where the theory of coherent algebraic sheaves
is constructed, for which the recently created theory of coherent analytic
sheaves served as a prototype (see § 8).
Generalizations of the concept of an algebraic variety, close in spirit
to the later concept of a scheme, were proposed at the beginning of
the 1950's. Apparently the first and for the time very systematic working
out of these ideas is due to Kahler [1], [2]. The concept of a scheme,
as well as the majority of results in the general theory of schemes, is
due to Grothendieck. The first systematic account of these ideas is
contained in a lecture by Grothendieck [1].
Bibliography for the Historical Sketch
Abel,N.H.: [1] Oeuvres completes. Christiania 1881.

Artin,E.: [1] Collected papers. New York-London: Addison-Wesley, 1965.
Bernoulli,J.: [1] Opera omnia, vol. I-IV. Lausannae et Genevae: Bosquet 1742.
Bertini,E.: [1] Ricerche sulle transformazioni univoche involutorie nel piano. Ann. Mat.
Pura Appl. (2) 8 (1877).
Brill,A., Noether,M.: [1] Uber die algebraischen Funktionen und ihre Anwendung in
der Geometrie. Math. Ann. 7 (1873).
Cartan,H.: [1] Varietes analytiques complexes et cohomologie, Coli. sur les fonctions de
plusieurs variables. Bruxelles, March 1953.
Castelnuovo,G.: [1] Sulla razionalita delle involuzioni piane. Rend. Accad. Lincei 2
(1893).
[2] Sulle superficie di genere zero. Mem. Soc. Ital. Sci. 10 (1896).
[3] Alcuni proprieta fondamentali dei sistemi lineari di curve tracciati sopra una
superficie, ibid.
[4] Sugli integrali semplici appartenenti and una superficie irregolars. Rend. Accad.
Lincei 14 (1905).
Clebsch,A.: [1] Sur les surfaces algebriques. C. R. Acad. Sci. Paris 67, 1238-1239 (1868).
Clebsch,A., Gordan, P.: [1] Theorie der Abelschen Funktionen. Leipzig: Teubner, 1866.
Dedekind,R., Weber, H.: [1] Theorie der algebraischen Funktionen einer Veranderlichen.
J. Reine Angew. Math. 92 (1882).
Enriques, F.: [1] Sulla proprieta caratheristica delle superficie irregolary. Rend. Accad.
Bologna 9 (1904).
[2] Superficie algebriche. Bologna 1949.
Euler,L.: [1] Integral calculus, Vol. I, Ch. VI.
[2] Opera omnia, Ser. I, Vol. XXI, 91-118.
Frobenius, G.: [1] Uber die Grundlagen der Theorie der Jakobischen Funktionen. J.
Reine Angew. Math. 97, 16-48, 188-223 (1884).
Gopel: [1] Theoriae transcendentium Abelianarum primi ordinis adumbrato levis. J.
Reine Angew. Math. 3S (1847).
Grothendieck,A.: [1] The cohomology theory of abstract algebraic varieties. Internat.
Congr. Math. Edinburgh 1958.
Hasse,H.: [1] Zur Theorie der abstrakten elliptischen Funktionenkorper. J. Reine Angew.
Math. 17S (1936).
[2] Zahlentheorie. Berlin: Akademie-Verlag, 1950.
Hopf,H.: [1] Zur Topologie der komplexen Mannigfaltigkeiten. Studies and essays
presented to R. Courant. New York 1958, 167-187.
Jacobi,C.G.J.: [1] Gesammelte Werke. Berlin 1881.
Kahler, E.: [1] Algebra und Differentialrechnung. Berichte Math. Tagung. Berlin 1953.
[2] Geometria arithmetica. Ann. Mat. Pura Appl. (4), 4S (1958).
432 Bibliography for the Historical Sketch
Klein, F.: [1] Gesammelte Mathematische Abhandlungen, Vol. III. Berlin: Springer-Verlag,
1923.
[2] Riemannsche Fliichen. Berlin 1891-1892.
Kronecker,L.: [1] Grundziige einer arithmetischen Theorie der algebraischen GraBen.
J. Reine Angew. Math. 92 (1882).
Lefschetz,S.: [1] Numerical invariants of algebraic varieties. Trans. Amer. Math. Soc. 22
(1921).
[2] L'Analysis situs et la geometrie algebrique. Paris 1924.
[3] Geometrie sur les surfaces et les varietes algebriques. Mem. Sci. Math. 40 (1929).
Legendre, A. M.: [1] Traite des fonctions elliptiques et des integrales Euleriennes, 3 vols.
Paris 1825-1828.
Noether,M.: [1] Zur Grundlegung der Theorie der algebraischen Raumkurven. J. Reine
Angew. Math. 93 (1882).
[2] Zur Theorie des eindeutigen Entsprechens algebraischer Gebilde von belie big
vielen Dimensionen. Math. Ann. 2 (1870), 8 (1875).
Picard,E.: [1] Sur les integrales des differentielles totales algebriques de premiere espece.
C. R. Acad. Sci. Paris 99 (1884).
Picard, E., Simart, G.: [1] Theorie des fonctions algebriques de deux variables independentes.
Paris 1897-1906.
Poincare, H.: [1] Oeuvres, Vol. II. Paris 1916.
[2] Sur les proprietes du potentiel et sur les fonctions Abeliennes. Acta Math. 22,
89-178 (1899).
Riemann,B.: [1] Gesammelte Werke.
Serre,J.P.: [1] Quelques problemes globaux relatifs aux varietes de Stein. ColI. sur les
fonctions de plusieurs variables. Bruxelles, March 1953.
[2] Faisceaux algebriques coherents. Ann. of Math. (2) 61 (1955).
Severi,F.: [1] La base minima pour la totalite des courbes algebriques tracees sur une
surface algebrique. Ann. Ecole Norm. Sup. 25 (1908).
Weil,A.: [1] Foundations of algebraic geometry. New York 1946.
[2] Sur la theorie des formes differentielles attachees a une variete analytique complexe.
Comm. Math. Helv. 24 (1947).
Weyl,H.: [1] Die Idee der Riemannschen Flache. Berlin 1923.
Subject Index
Addition of points on plane cubic curve Complete intersection 401

148 - variety 265
Affine variety 36 Completion of module 96
Algebraic curve, branch 123 Complex dimension 345
- -,genus 171 - torus 347
- -, irreducible 5 Convergence in ring of formal power series
- -, moduli 178 94
- -, plane 3 Coordinate ring of a closed subset of an
- -, rational 5 affine space 16
- -, real 336 Cotangent bundle 274
- family of divisors 154 Criterion of Chevalley-Kleiman 292
- group 150 - of Grauert 293
Analytic space 357 - ofNakai-Moishezon 293
Automorphic form 393 - for being ruled 180
- function 399 - for rationality 180
Curve, hyperelliptic 175
Bezout's theorem in projective space 199 Cycle on algebraic variety 206
- - in product of projective spaces 199
- - on curve 145 Decomposition, incontractible, of closed
Bundle, corresponding to divisor 278 set into irreducibles 23
-,cotangent 274 Degree of divisor on curve 140
-, linear 277 - of mapping 116
-, normal 275 Dense set 15
Differential form, invariant 168
Canonical class 170 - of a function 156
Centre of u-process 104 - - - -- at a point 75
Characteristic class of a linear bundle 279 Dimension of divisor 137
Chow coordinates of a variety 66 -, complex 345
Chow's lemma 282 - of complex space 357
Class, canonical 170 - of q uasiprojective variety 53
-, characteristic of a linear bundle 279 - of ring 232
- of divisors 131 - of topological space 232
- - - of degree zero on an algebraic Direct sum of sheaves 272
curve 145
- - of vector bundles 273
Closed embedding of schemes 248 Divisor 127
- subscheme 248 -, effective 127, 360
- subvariety 264 -, locally principal 133
Closure of an ideal 96 - ofa form 133
Codimension of subvariety 53 - of a function 129
Combinatorial surface 334 - of differential form 170
434 Subject Index
Divisor of hyperplane section 133 Group of divisor classes 131

- of meromorphic function 363 -, Picard 133
- of poles of a function 129
- of zeros of a function 129 Harnack's theorem 337
- on analytic manifold 360 Hilbert's basis theorem XIII
- class 131 - Nullstellensatz XIII
Holomorphically convex space 407
Elliptic integral 391 - - -, complete 407
Embedding of schemes, closed 248 Homomorphism of sheaves of modules
Equivalence of divisors 131 271
-, algebraic 154 Hopf manifold 348
- -, of cycles 207 Hyperelliptic curve 175
-, numerical 288 Hypersurface in an affine space 16
Exceptional subvariety 105 - as closed subset 18
-, projective 40
Factor bundle 275
- sheaf 296
Ideal of closed subset in the ring of regular
Factorial variety 112
functions on large closed subset of an
Family, algebraic, of divisors 154
affine space 18
-, -, of varieties 68
- - - - of an affine space 16
- of vector spaces 268
- - - - of a projective space 30
Field of rational functions on quasipro-
Image of closed set under rational mapping
jective variety 38
26
- - - - as irreducible closed subset
- of homomorphism of sheaves 295
of affine space 24
Incontractible decomposition of closed
- - - - of plane irreducible algebraic
set into irreducibles 23
curve 8
- - - - of variety 263 Integral, elliptic 391
Intersection index of divisors 195
Finiteness theorem 304
- - - - in general position 183
Flat ring 96
- - of a curve and a divisor 287
Form, automorphic 393
- - of cycles on a differentiable mani-
-, differential, invariant 168
-, -, domain of regularity 164 fold 317
- - of effective divisors at a point 182
-, -, rational 164
In variant differential form 168
-, -, regular one-dimensional 156
Inverse image, proper, under cr-process
-, -, -, r-dimensional 161
Fraction, meromorphic 360 209
- -, -, of subvarieties 287
Frobenius relations 356
- - of scheme 250
Function, analytic on complex variety 345
Isomorphism of birational quasiprojective
-, automorphic 399
varieties 39
-, elliptic 390
-, birational of closed subsets of an affine
-, meromorphic 362
space 26
-, rational, domain of definition 25
-, - of plane irreducible algebraic curves
-, -, regular at point 24
11
-, regular at point 33
- of closed subsets of an affine space
-, -, on closed subset of affine space 16
20
Fundamental polygon 400
- of quasiprojective varieties 36
General position of divisors 182
Generic point 230 Jacobian variety 155
Genus of algebraic curve 171
Group, algebraic 150 Kernel of homomorphism of sheaves 295
-, modular 392 -, local, of cr-process 101
Subject Index 435
Kernel of 6-function 387 Multiplicity of singular point of a plane

- with centre in smooth subvariety 284 curve 80, 185
- - - at point of quasiprojective - of intersection of divisors in a sub-
variety 103 variety 190
- - - - of projective space 99 - - - of effective divisors at a point
182
Lattice 347
Linear bundle 277 Nakayama's lemma 83
- divisors, system of 138 Neighbourhood of point 15
Local equations of subvariety 90 - - -, infinitely small 251
- parameters at a point 81 Nil-radical 227
- ring of an irreducible subvariety 188 Normal bundle 275
- - of point 72 Normalization of quasiprojective variety
- - - - of a spectrum 227 113
- - of prime ideal 71 - of varieties in finite extension of func-
Liiroth's problem 174 tion field 266
- theorem 9 - theorem 52
Noetherian scheme 252
Manifold, analytic 344 Order of ramification of mapping of curves
-,-,complex 344 327
-, -, of elliptic type 344 Ordinary singular point 123
-, -, of hyperbolic type 382 Orientation of differentiable manifold 312
-, -, of parabolic type 344 - of canonical manifold X (CC) 312
-, Hopf 348 - of triangulation 335
-, -, generalized 378 Ovals of real curve 340
-, orienting class of 313
Mapping, finite of affine varieties 48 Parameters, local, at a point 81
-, -, of quasiprojective varieties 49 Pasting of schemes 246
-, holomorphic 345 Picard group 133
- of spectra associated with homomor- Poincare series 393
phism of rings 224 Poincare's theorem on complete irre-
-, rational, of irreducible subset of affine ducibility 351
space 25 Point, generic 230
- -, of plane irreducible algebraic curves Polygon, fundamental 400
11 Presheaf 234
-, -, regular at a point 26 - of regular functions 235
-, regular, of closed subjects of affine -, structure 235
space 18 Principal open set 37
-, -, of quasiprojective varieties 34 - - - of spectrum 229
-, restriction, of sheaf 235 Product of closed subsets of affine space
-, unramified regular at a point 117 16
-, Veronese 40 - of cycles 206
Meromorphic fraction 360 - of schemes 256
Modeloffield 106 Projection with centre in subspace 40
-, relatively minimal 107 Projective space over a ring 247
Module, completion 96 - variety 36
-, length 189 Projectization of vector bundle 285
-, of differentials of ring 160 Quadratic transformation 215
Morphism of families of vector spaces 268 Quasiprojective variety 33
- of ringed spaces 242
- of varieties 263 Ramification point 117
-, rational, of schemes 246 Rational differential form 164
436 Subject Index
Rational variety 174 Subset, affine open of closed subset of

Reducibility of topological space 230 projective space 32
Reducible Space 230 -, - - of projective space 14
Regular vector field 305 -, closed, of affme space 14
Restriction mapping of sheaf 235 -, -, irreducible 22
- of divisor 134 -, -, of projective space 31
Riemann-Roch theorem 176 -, -, of quasiprojective variety 33
Riemann's existence theorem 351 - -, reducible 22
Ring of formal power series 84 -, open, of affine space 15
Ringed space 242 -, -, of closed subset of projective space
Ruled surface 108 32
-, -, of quasiprojective variety 33
Scheme 244 Subsheaf 295
-, affine 245 Subspace of complex space 358
-, defined over a field 310 Subvariety, closed 264
-, diagonal 258 -, codimension of 53
-, group 258 -, expectional 105
-, Noetherian 252 -, local equations 90
- of finite type 252 - of quasiprojective variety 33
- over ring 244 Support of divisor 127
-, reduced 250 - - - on analytic manifold 361
-, separated 258 - of locally principal divisor 133
Section of vector bundle 270 - of sheaf 297
Set, dense 15 Surface, combinatorial 334
Sheaf 238 -, elliptic 179
-, associated with a presheaf 242 -, Euler, characteristics 336
-,coherent 299 - of type K3 179
-, corresponding to vector bundle 272 - - - -, analytic 378
-, - to divisor 278 -, ruled 108
-, dual 272
-, exterior power 272
-, invertible 279 Tangency of a line and a variety 73
-, locally free 272 Tangent bundle of quasiprojective variety
- of analytic functions 345 77
- of differential forms 273 - - of variety 274
- of fibres 241 - cone 79
- of modules over a sheaf of rings 272 - space at point of spectrum 227
-, structure, of ringed space 243 - - of variety 73
Simple sequence 186 Taylor series of a function 85
Singular point with distinct tangents 123 Tensor product of vector bundles 273
Space associated with divisor 137 - - of sheaves 272
-, projective over a ring 247 Theorem of Siegel 365
-, reducible 230 - on closure of projective image of variety
Spectrum of ring 224 44
- - -, maximal 223 - - dimension of intersection with
-, regular point 227 hypersurface 57
-, simple point 227 - - - of fibres of mapping 60
-, specialization of point 230 Topology complex, of quasiprojective
Structure presheaf 235 varieties 89
- sheaf of ringed space 243 - real, of quasiprojective varieties 89
Subbundle 274 -, spectral 228
Subscheme, closed 248 Transition matrix of vector bundle 270
Subject Index 437
Transversality of subvarieties 82 Variety, normal 109

Triangulation 333 -, Picard 155
-, projective 36
-, quasiprojective 33
Uniformization of curve 390 -, rational 174
U nirational variety 174 -, simple point 79
Unramified covering of topological space -, singular point 78
346 -, smooth 77
-, unrational 174
Variety 263 -, Veronese 40
-, Abelian 152 Vector bundle 268
-, affine 36 - -,dual 272
-, associated form 66 - -, exterior power 272
-, complete 265 Veronese mapping 40
-, factorial 112 - variety 40
-, Jacobian 155
-, non-singular of codimension 1 90 Weierstrass' preparation theorem 95
List of Notation
IAn n-dimensional affne space 14

XxY product of affine varieties X and Y 16
product of quasiprojective varieties X and Y 42
product of schemes X and Y 255
k[XJ coordinate ring of closed subject X of affine space 16
ring of regular functions on quasiprojective variety X 34
Illx ideal of closed subset X of affine space 16
ideal of closed subset X of projective space 31
cry ideal of closed subset Y in ring of functions on closed set X) Y 18
V(u) hypersurface with equation u(x) = 0 in closed subset of affine space 18
subset of spectrum determined by the element u 229
f* mapping of function corresponding to regular mapping f 19
mapping of function corresponding to rational mapping f 27
mapping of divisors corresponding to regular mapping f 134
Ll diagonal of varieties 21
diagonal of schemes 258
k(X) field of rational functions on closed subset X of affine space 24
field of rational functions on quasi projective variety X 38
field of rational functions on variety X 263
!pn n-dimensional projective space 30
D(f) principal open set in quasiprojective variety 37
principal open set in scheme 229
Vm Veronese mapping 40
dimX dimension of variety X 53
codimxY codimension of subvariety Y in X 53
A. local ring of prime ideal p 71
(!Jx local ring of point x of variety 72
local ring of point x of spectrum 227
ex tangent space at point x of variety 73
tangent space at point x of spectrum 227
dx(f) differential of function f at point x 75
k[[TJJ ring of formal power series 85
X' normalization of quasiprojective variety X 113
normalization of variety X 226
v normalization mapping of quasiprojective variety 113
normalization mapping of variety 266
List of Notation 439
deg(f) degree of mapping f 116

(f) divisor of function f 129
equivalence of divisors 131
Div(X) group of divisors 130
CI(X) group of divisor classes 130
(F) divisor of form F 133
Pic (X) Picard group 133
£,(D) space associated with divisor D 137
/(D) dimension of divisor D 137
degD degree of divisor D on curve 140
df differential of function f 156
Q"[X] module of r-dimensional regular differential forms on X 161
Q"(X) space of r-dimensional rational differential forms on X 164
hr = dimQ" [X] 168
(w) divisor of differential form w 170
Kx canonical class of X 170
g(X) genus of curve X 171
(D1 , ..• ,Dn)x intersection index of divisors D1 , ••. , Dn at point x 183
(D1 , ..• , Dn)e intersection index of divisors D1 , ..• , Dn in variety C 190
(Dl,···,Dn) intersection index of divisors D1 , ..• , Dn 183, 195
Spec A spectrum of ring A 224
"cp mapping of spectra associated with homomorphism cp 224
As localization of ring A with respect to mUltiplicative system S 225
Af localization of ring A with respect to multiplicative system {fn} 226
eb,eb.F mappings defining a sheaf 234
(!J structure presheaf (shea!) on spectrum of ring 235
Fx fibre of sheaf F at point x 241
(!Jx structure sheaf of ringed space 243
£'(E) space of sections of vector bundle E 271
£IE sheaf corresponding to bundle E 272
detE determinant of bundle E 273
e tangent bundle 274
N x/y normal bundle 275
£'D sheaf corresponding to divisor D 277
ED bundle corresponding to divisor D 277
X(CC) space of closed points in complex topology 309
WM orienting class of manifold M 313
(!J.n sheaf of analytic functions 344
fan holomorphic mapping corresponding to morphism f 345
.R(X) field of meromorphic functions on analytic manifold X 363
O(z) theta function 387

Basic Algebraic Geometry

Uploaded by

Copyright:

Available Formats

Basic Algebraic Geometry

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Basic Algebraic Geometry

Uploaded by

Copyright:

Available Formats

Grundlehren der

mathematischen Wissenschaften 213

Translated from the Russian by K. A. Hirsch

Title of the Russian Original Edition: Osnovy algebraicheskoi geometrii,

AMS Subject Classifications (1970): Primary 14-XX, Secondary 20 GXX

ISBN-13: 978-3-540-08264-4 e-ISBN-13: 978-3-642-96200-4

Algebraic geometry occupied a central place in the mathematics of the

to the disappearance of that vivid geometric style that was charac-

The aim of the book is to set forth the elements of algebraic

Part I. Algebraic Varieties in a Projective Space

Chapter II. Local Properties

§ 3. Properties of Simple Points 90

Chapter III. Divisors and Differential Forms

§ 2. Divisors on Curves 140

§ 3. Algebraic Groups . . . . . . . . . . . . 148

§ 4. Differential Forms . . . . . . . . . . . . . . . . 156

§ 5. Examples and Applications of Differential Forms 166

Chapter IV. Intersection Indices

§ 2. Applications and Generalizations of Intersection Indices . . . . . 198

§ 3. Birational Isomorphisms of Surfaces 208

Part II. Schemes and Varieties

Chapter VI. Varieties

§ 3. Coherent Sheaves 294

Part III. Algebraic Varieties over the Field of Complex

§ 3. The Topology of Algebraic Curves 325

§ 4. Real Algebraic Curves 336

Chapter VIII. Complex Analytic Manifolds

Chapter IX. Uniformization

§ 2. Curves of Parabolic Type 386

§ 3. Curves of Hyperbolic Type . 393

§ 4. On the Uniformization of Manifolds of Large Dimension . 401

The reader who is interested only in vanetles over the field of

In our book the number-theoretical aspect of algebraic geometry

§ 1. Plane Algebraic Curves

1. Rational Curves. The curve given by the equation

A plane algebraic curve is the set of all points whose coordinates

and X4 + y4 = 0 determine one and the same "curve" - the ongm of

In all these cases the investigation of points whose coordinates lie

The problem of the existence of at least one solution is fairly

rational function. It is easy to check that we arrive in this way at

where p, q and r are polynomials, which we may assume to be relatively

must hold identically. Differentiating it and dividing the result by n

we obtain the relation

We denote the degrees of p, q, and r by a, b, and c, with a;;.. b;;.. c, say.

2. Connections with the Theory of Fields. We now show that

variable x, therefore the field of rational functions on a line is the same

transcendence degree of Kover k is 1, x and y must be connected by an

x = cp(X(x, y»), y= tp (X (x, y»), ( 1)

t = X(cp(t), tp(t») . (2)

Observe that we have proved 1) and 2) not for any parametrization

3. Birational Isomorphism of Curves. If a plane algebraic curve

Two curves X and Yare called birationally isomorphic if there exist

As in the investigation of rational curves, we can establish a connection

We see that the problem of classifying algebraic curves to within

y2= n2n (X-CXi) (X-CXi- l )

into the curve

Is this mapping a birational isomorphism?

§ 2. Closed Subsets of Affine Spaces

1. Definition of Closed Subset. At various stages of the develop-

The union of rmitely many closed sets is also closed. Clearly, it is